npm - @blockrun/clawrouter - Versions diffs - 0.12.227 → 0.12.229 - Mend

@blockrun/clawrouter 0.12.227 → 0.12.229

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (7) hide show

package/README.md +53 -52
package/dist/cli.js +83 -55
package/dist/index.js +83 -55
package/package.json +1 -1
package/skills/clawrouter/SKILL.md +2 -2
package/skills/imagegen/SKILL.md +9 -9
package/src/top-models.json +2 -2

package/README.md CHANGED Viewed

@@ -207,12 +207,12 @@ Choose your routing strategy with `/model <profile>`:
 Request → Weighted Scorer (15 dimensions) → Tier → Best Model → Response
 ```
-| Tier      | ECO Model                             | AUTO Model                            | PREMIUM Model                |
-| --------- | ------------------------------------- | ------------------------------------- | ---------------------------- |
-| SIMPLE    | free/mistral-large-3-675b (**FREE**)  | gemini-2.5-flash ($0.30/$2.50)        | kimi-k2.7 ($0.95/$4.00)      |
-| MEDIUM    | gemini-3.1-flash-lite ($0.25/$1.50)   | kimi-k2.7 ($0.95/$4.00)               | gpt-5.3-codex ($1.75/$14.00) |
-| COMPLEX   | gemini-3.1-flash-lite ($0.25/$1.50)   | gemini-3.1-pro ($2/$12)               | claude-opus-4.8 ($5/$25)     |
-| REASONING | deepseek-reasoner ($0.20/$0.40)       | deepseek-reasoner ($0.20/$0.40)       | claude-sonnet-4.6 ($3/$15)   |
+| Tier      | ECO Model                            | AUTO Model                      | PREMIUM Model                |
+| --------- | ------------------------------------ | ------------------------------- | ---------------------------- |
+| SIMPLE    | free/mistral-large-3-675b (**FREE**) | gemini-2.5-flash ($0.30/$2.50)  | kimi-k2.7 ($0.95/$4.00)      |
+| MEDIUM    | gemini-3.1-flash-lite ($0.25/$1.50)  | kimi-k2.7 ($0.95/$4.00)         | gpt-5.3-codex ($1.75/$14.00) |
+| COMPLEX   | gemini-3.1-flash-lite ($0.25/$1.50)  | gemini-3.1-pro ($2/$12)         | claude-opus-4.8 ($5/$25)     |
+| REASONING | deepseek-reasoner ($0.20/$0.40)      | deepseek-reasoner ($0.20/$0.40) | claude-sonnet-4.6 ($3/$15)   |
 **Blended average: $2.05/M** vs $25/M for Claude Opus = **92% savings**
@@ -230,16 +230,16 @@ Generate images directly from chat with `/cr-imagegen`:
 > The slash command is `/cr-imagegen` to avoid colliding with Telegram channel commands. Typing `/imagegen` in chat still works for backward compatibility.
-| Model                        | Provider              | Price        | Max Size  |
-| ---------------------------- | --------------------- | ------------ | --------- |
-| `nano-banana`                | Google Gemini Flash   | $0.05/image  | 1024x1024 |
-| `banana-pro`                 | Google Gemini Pro     | $0.10/image  | 4096x4096 |
-| `gpt-image`                  | OpenAI GPT Image 1    | $0.02/image  | 1536x1024 |
-| `gpt-image-2`                | OpenAI GPT Image 2    | $0.06/image  | 1536x1024 |
-| `seedream`                   | ByteDance Seedream 5  | $0.045/image | 2848x1600 |
-| `xai/grok-imagine-image`     | xAI Grok Imagine      | $0.02/image  | 1024x1024 |
-| `xai/grok-imagine-image-pro` | xAI Grok Imagine Pro  | $0.07/image  | 1024x1024 |
-| `zai/cogview-4`              | Zhipu CogView-4       | $0.015/image | 1440x1440 |
+| Model                        | Provider             | Price        | Max Size  |
+| ---------------------------- | -------------------- | ------------ | --------- |
+| `nano-banana`                | Google Gemini Flash  | $0.05/image  | 1024x1024 |
+| `banana-pro`                 | Google Gemini Pro    | $0.10/image  | 4096x4096 |
+| `gpt-image`                  | OpenAI GPT Image 1   | $0.02/image  | 1536x1024 |
+| `gpt-image-2`                | OpenAI GPT Image 2   | $0.06/image  | 1536x1024 |
+| `seedream`                   | ByteDance Seedream 5 | $0.045/image | 2848x1600 |
+| `xai/grok-imagine-image`     | xAI Grok Imagine     | $0.02/image  | 1024x1024 |
+| `xai/grok-imagine-image-pro` | xAI Grok Imagine Pro | $0.07/image  | 1024x1024 |
+| `zai/cogview-4`              | Zhipu CogView-4      | $0.015/image | 1440x1440 |
 ## Video Generation
@@ -382,44 +382,45 @@ No Surf account, no API key — settles directly to Surf's Base treasury in USDC
 ## Models & Pricing
-55+ models across 9 providers, one wallet. **Starting at $0.0002/request.**
+55+ models across 9 providers, one wallet. **8 models are $0 — paid models start at fractions of a cent.**
-> **💡 "Cost per request"** = estimated cost for a typical chat message (~500 input + 500 output tokens).
+> **💡 "Cost per request"** = estimated cost for a typical chat message (~500 input + 500 output tokens). Paid requests also carry a flat **$0.002/tx settlement fee** (covers on-chain gas; already included in the price the gateway quotes). Free models never pay it.
 ### Budget Models (under $0.001/request)
-| Model                                         | Input $/M | Output $/M | ~$/request | Context | Features                                     |
-| --------------------------------------------- | --------: | ---------: | ---------: | ------- | -------------------------------------------- |
-| free/mistral-large-3-675b                     |  **FREE** |   **FREE** |     **$0** | 131K    | 675B general flagship, default free model    |
-| free/qwen3-next-80b-a3b-instruct              |  **FREE** |   **FREE** |     **$0** | 262K    | reasoning + coding                           |
-| free/seed-oss-36b                             |  **FREE** |   **FREE** |     **$0** | 131K    | coding                                       |
-| free/nemotron-3-nano-omni-30b-a3b-reasoning   |  **FREE** |   **FREE** |     **$0** | 256K    | reasoning, **vision** (text+img+video+audio) |
-| free/mistral-nemotron                         |  **FREE** |   **FREE** |     **$0** | 131K    | instruction following                        |
-| free/step-3.7-flash                           |  **FREE** |   **FREE** |     **$0** | 131K    | reasoning                                    |
-| free/nemotron-nano-9b-v2                      |  **FREE** |   **FREE** |     **$0** | 131K    | fast lightweight generalist                  |
-| free/nemotron-nano-12b-v2-vl                  |  **FREE** |   **FREE** |     **$0** | 131K    | vision (text + image)                        |
-| openai/gpt-5-nano                             |     $0.05 |      $0.40 |    $0.0002 | 128K    | tools                                        |
-| openai/gpt-4.1-nano                           |     $0.10 |      $0.40 |    $0.0003 | 128K    | tools                                        |
-| google/gemini-2.5-flash-lite                  |     $0.10 |      $0.40 |    $0.0003 | 1M      | tools                                        |
-| openai/gpt-4o-mini                            |     $0.15 |      $0.60 |    $0.0004 | 128K    | tools                                        |
-| xai/grok-4-fast                               |     $0.20 |      $0.50 |    $0.0004 | 131K    | tools                                        |
-| xai/grok-4-fast-reasoning                     |     $0.20 |      $0.50 |    $0.0004 | 131K    | reasoning, tools                             |
-| xai/grok-4-1-fast                             |     $0.20 |      $0.50 |    $0.0004 | 131K    | tools                                        |
-| openai/gpt-5-mini                             |     $0.25 |      $2.00 |    $0.0011 | 200K    | tools                                        |
-| deepseek/deepseek-chat                        |     $0.20 |      $0.40 |    $0.0003 | 1M      | tools (V4 Flash chat)                        |
-| deepseek/deepseek-reasoner                    |     $0.20 |      $0.40 |    $0.0003 | 1M      | reasoning, tools (V4 Flash thinking)         |
-| deepseek/deepseek-v4-pro                      |    $0.435 |      $0.87 |    $0.0007 | 1M      | reasoning, agentic, tools (V4 flagship)      |
-| zai/glm-5                                     |     $0.60 |      $1.92 |    $0.0013 | 200K    | tools                                        |
-| zai/glm-5-turbo                               |     $1.20 |      $4.00 |    $0.0026 | 200K    | tools                                        |
-| minimax/minimax-m3                            |     $0.30 |      $1.20 |    $0.0008 | 1M      | reasoning, agentic, tools                    |
-| minimax/minimax-m2.7                          |     $0.30 |      $1.20 |    $0.0008 | 205K    | reasoning, agentic, tools                    |
-| minimax/minimax-m2.5                          |     $0.30 |      $1.20 |    $0.0008 | 205K    | reasoning, agentic, tools                    |
-| google/gemini-2.5-flash                       |     $0.30 |      $2.50 |    $0.0014 | 1M      | vision, tools                                |
-| openai/gpt-4.1-mini                           |     $0.40 |      $1.60 |    $0.0010 | 128K    | tools                                        |
-| google/gemini-3.5-flash                       |     $0.50 |      $3.00 |    $0.0018 | 1M      | reasoning, vision, tools (thinking built-in) |
-| google/gemini-3-flash-preview                 |     $0.50 |      $3.00 |    $0.0018 | 1M      | vision                                       |
-| moonshot/kimi-k2.5                            |     $0.60 |      $3.00 |    $0.0018 | 262K    | reasoning, vision, agentic, tools            |
-| moonshot/kimi-k2.7                            |     $0.95 |      $4.00 |    $0.0025 | 262K    | reasoning, vision, agentic, tools            |
+| Model                                       | Input $/M | Output $/M | ~$/request | Context | Features                                     |
+| ------------------------------------------- | --------: | ---------: | ---------: | ------- | -------------------------------------------- |
+| free/mistral-large-3-675b                   |  **FREE** |   **FREE** |     **$0** | 131K    | 675B general flagship, default free model    |
+| free/deepseek-v4-flash                      |  **FREE** |   **FREE** |     **$0** | 1M      | reasoning + coding, huge context             |
+| free/seed-oss-36b                           |  **FREE** |   **FREE** |     **$0** | 131K    | coding                                       |
+| free/nemotron-3-nano-omni-30b-a3b-reasoning |  **FREE** |   **FREE** |     **$0** | 256K    | reasoning, **vision** (text+img+video+audio) |
+| free/mistral-nemotron                       |  **FREE** |   **FREE** |     **$0** | 131K    | instruction following                        |
+| free/step-3.7-flash                         |  **FREE** |   **FREE** |     **$0** | 131K    | reasoning                                    |
+| free/nemotron-nano-9b-v2                    |  **FREE** |   **FREE** |     **$0** | 131K    | fast lightweight generalist                  |
+| free/nemotron-nano-12b-v2-vl                |  **FREE** |   **FREE** |     **$0** | 131K    | vision (text + image)                        |
+| openai/gpt-5-nano                           |     $0.05 |      $0.40 |    $0.0002 | 128K    | tools                                        |
+| openai/gpt-4.1-nano                         |     $0.10 |      $0.40 |    $0.0003 | 128K    | tools                                        |
+| google/gemini-2.5-flash-lite                |     $0.10 |      $0.40 |    $0.0003 | 1M      | tools                                        |
+| openai/gpt-4o-mini                          |     $0.15 |      $0.60 |    $0.0004 | 128K    | tools                                        |
+| xai/grok-4-fast                             |     $0.20 |      $0.50 |    $0.0004 | 131K    | tools                                        |
+| xai/grok-4-fast-reasoning                   |     $0.20 |      $0.50 |    $0.0004 | 131K    | reasoning, tools                             |
+| xai/grok-4-1-fast                           |     $0.20 |      $0.50 |    $0.0004 | 131K    | tools                                        |
+| openai/gpt-5-mini                           |     $0.25 |      $2.00 |    $0.0011 | 200K    | tools                                        |
+| deepseek/deepseek-chat                      |     $0.20 |      $0.40 |    $0.0003 | 1M      | tools (V4 Flash chat)                        |
+| deepseek/deepseek-reasoner                  |     $0.20 |      $0.40 |    $0.0003 | 1M      | reasoning, tools (V4 Flash thinking)         |
+| deepseek/deepseek-v4-pro                    |    $0.435 |      $0.87 |    $0.0007 | 1M      | reasoning, agentic, tools (V4 flagship)      |
+| zai/glm-5                                   |     $0.60 |      $1.92 |    $0.0013 | 200K    | tools                                        |
+| zai/glm-5-turbo                             |     $1.20 |      $4.00 |    $0.0026 | 200K    | tools                                        |
+| minimax/minimax-m3                          |     $0.30 |      $1.20 |    $0.0008 | 1M      | reasoning, agentic, tools                    |
+| minimax/minimax-m2.7                        |     $0.30 |      $1.20 |    $0.0008 | 205K    | reasoning, agentic, tools                    |
+| minimax/minimax-m2.5                        |     $0.30 |      $1.20 |    $0.0008 | 205K    | reasoning, agentic, tools                    |
+| google/gemini-2.5-flash                     |     $0.30 |      $2.50 |    $0.0014 | 1M      | vision, tools                                |
+| openai/gpt-4.1-mini                         |     $0.40 |      $1.60 |    $0.0010 | 128K    | tools                                        |
+| google/gemini-3.5-flash                     |     $0.50 |      $3.00 |    $0.0018 | 1M      | reasoning, vision, tools (thinking built-in) |
+| google/gemini-3-flash-preview               |     $0.50 |      $3.00 |    $0.0018 | 1M      | vision                                       |
+| moonshot/kimi-k2.5                          |     $0.60 |      $3.00 |    $0.0018 | 262K    | reasoning, vision, agentic, tools            |
+| moonshot/kimi-k2.7                          |     $0.95 |      $4.00 |    $0.0025 | 262K    | reasoning, vision, agentic, tools            |
+| moonshot/kimi-k3                            |     $3.00 |     $15.00 |    $0.0110 | 1M      | reasoning, vision, agentic, tools            |
 ### Mid-Range Models ($0.001–$0.01/request)
@@ -461,7 +462,7 @@ No Surf account, no API key — settles directly to Surf's Base treasury in USDC
 | openai/gpt-5.2-pro          |    $21.00 |    $168.00 |    $0.0945 | 400K    | reasoning, tools                  |
 | openai/gpt-5.4-pro          |    $30.00 |    $180.00 |    $0.1050 | 400K    | reasoning, tools                  |
-> **Free tier:** several NVIDIA-hosted models cost nothing — `/model free` smart-routes across them, or pick one directly (e.g., `/model nemotron-omni` for vision, `/model qwen3-next` for 262K reasoning, `/model seed-oss` for coding, `/model mistral-large` for the 675B flagship).
+> **Free tier:** several NVIDIA-hosted models cost nothing — `/model free` smart-routes across them, or pick one directly (e.g., `/model nemotron-omni` for vision, `/model v4-flash` for 1M-context reasoning, `/model seed-oss` for coding, `/model mistral-large` for the 675B flagship).
 > **Best value:** `gpt-5-nano` and `gemini-2.5-flash-lite` deliver strong results at ~$0.0003/request.
 ---
@@ -720,7 +721,7 @@ ClawRouter works with any tool that makes OpenAI-compatible API calls — point
 ### Is ClawRouter free?
-ClawRouter itself is free and MIT licensed. You pay only for the LLM API calls routed through it — and several NVIDIA-hosted models (`mistral-large-3-675b`, `qwen3-next-80b-a3b-instruct`, `seed-oss-36b`, `nemotron-3-nano-omni-30b-a3b-reasoning`, `mistral-nemotron`, `step-3.7-flash`, `nemotron-nano-9b-v2`, `nemotron-nano-12b-v2-vl`) are completely free. Use `/model free` to smart-route across them, or pick any by name.
+ClawRouter itself is free and MIT licensed. You pay only for the LLM API calls routed through it — and several NVIDIA-hosted models (`mistral-large-3-675b`, `deepseek-v4-flash`, `seed-oss-36b`, `nemotron-3-nano-omni-30b-a3b-reasoning`, `mistral-nemotron`, `step-3.7-flash`, `nemotron-nano-9b-v2`, `nemotron-nano-12b-v2-vl`) are completely free. Use `/model free` to smart-route across them, or pick any by name.
 ---

package/dist/cli.js CHANGED Viewed

@@ -33992,10 +33992,10 @@ var init_config = __esm({
             // FREE — smaller, faster
             "free/mistral-large-3-675b",
             // FREE — 675B general flagship (re-featured 2026-06-14)
-            "free/qwen3.5-122b-a10b",
-            // FREE — newest-gen Qwen, strong general
-            "free/qwen3-next-80b-a3b-instruct",
-            // FREE — 262K ctx, strong general (deepseek-v4-flash retired)
+            "free/deepseek-v4-flash",
+            // FREE — 1M ctx, recovered in blockrun's 2026-07-17 re-probe
+            "free/seed-oss-36b",
+            // FREE — live coder (qwen3.5-122b + qwen3-next died 2026-07)
             "google/gemini-3.1-flash-lite",
             // $0.25/$1.50 — newest flash-lite
             "openai/gpt-5.4-nano",
@@ -34295,12 +34295,12 @@ var init_top_models = __esm({
       "zai/glm-5",
       "minimax/minimax-m3",
       "minimax/minimax-m2.7",
-      "moonshot/kimi-k2.7",
+      "moonshot/kimi-k3",
       "deepseek/deepseek-v4-pro",
       "deepseek/deepseek-chat",
       "deepseek/deepseek-reasoner",
       "free/mistral-large-3-675b",
-      "free/qwen3-next-80b-a3b-instruct",
+      "free/deepseek-v4-flash",
       "free/seed-oss-36b",
       "free/nemotron-3-nano-omni-30b-a3b-reasoning",
       "free/mistral-nemotron",
@@ -34507,13 +34507,16 @@ var init_models = __esm({
       deepseek: "deepseek/deepseek-chat",
       "deepseek-chat": "deepseek/deepseek-chat",
       reasoner: "deepseek/deepseek-reasoner",
-      // Kimi / Moonshot — K2.7 is the featured flagship on BlockRun (added 2026-06-13,
-      // commit cd3d79b; K2.6 marked hidden/superseded, K2.5 hidden). Bare aliases now
-      // resolve to K2.7 (same $0.95/$4.00 price as K2.6 — AT-COST, zero margin). Explicit
+      // Kimi / Moonshot — K3 is the featured flagship on BlockRun (added 2026-07-17; K2.7
+      // hidden/superseded, K2.6 hidden, K2.5 hidden). K3 is ~5x K2.7's price ($3/$15 vs
+      // $0.95/$4.00), so the BARE aliases deliberately STAY on K2.7 — repointing them to K3
+      // would silently ~5x every generic-`kimi` quote and break per-call-cap wallets (mirrors
+      // blockrun's own alias decision). Address the flagship explicitly via "kimi-k3". Explicit
       // pins for "kimi-k2.6" / "kimi-k2.5" still resolve to those exact models (K2.5 is a
       // cost-stability opt-in at $0.60/$3.00). NVIDIA-hosted K2.5 was retired 2026-04-21.
       kimi: "moonshot/kimi-k2.7",
       moonshot: "moonshot/kimi-k2.7",
+      "kimi-k3": "moonshot/kimi-k3",
       "kimi-k2": "moonshot/kimi-k2.7",
       "kimi-k2.7": "moonshot/kimi-k2.7",
       "kimi-k2.6": "moonshot/kimi-k2.6",
@@ -34570,14 +34573,12 @@ var init_models = __esm({
       "gpt-20b": "free/gpt-oss-20b",
       "nvidia/gpt-oss-120b": "free/gpt-oss-120b",
       "nvidia/gpt-oss-20b": "free/gpt-oss-20b",
+      // deepseek family → v4-flash, matching blockrun's 2026-07-17 redirect map
+      // (v4-flash recovered in the live re-probe; v3.2/v4-pro stay dead upstream).
       "nvidia/deepseek-v3.2": "free/deepseek-v4-flash",
-      // v3.2 → v4-flash (v4-pro NVIDIA hung 2026-04-30)
       "free/deepseek-v3.2": "free/deepseek-v4-flash",
-      // local pin redirect
       "nvidia/deepseek-v4-pro": "free/deepseek-v4-flash",
-      // V4 Pro NVIDIA hung 2026-04-30
       "free/deepseek-v4-pro": "free/deepseek-v4-flash",
-      // V4 Pro delisted — redirect pinned callers
       "nvidia/deepseek-v4-flash": "free/deepseek-v4-flash",
       "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning": "free/nemotron-3-nano-omni-30b-a3b-reasoning",
       // qwen3-coder-480b retired (NVIDIA EOL 2026-06-14) → redirects to seed-oss-36b
@@ -34585,58 +34586,62 @@ var init_models = __esm({
       // resolves them), but point the generic coding shorthands at the live successor.
       "nvidia/qwen3-coder-480b": "free/qwen3-coder-480b",
       "qwen/qwen3-coder-480b-a35b-instruct": "free/qwen3-coder-480b",
-      "nvidia/glm-4.7": "free/seed-oss-36b",
+      // glm-4.7 pin keeps the real id — blockrun redirects it server-side (→ gpt-oss-120b).
+      "nvidia/glm-4.7": "free/glm-4.7",
       "nvidia/llama-4-maverick": "free/llama-4-maverick",
-      // qwen3-next thinking variant retired; live sibling is the instruct model
+      // qwen3-next: both variants died in the 2026-07-17 re-probe; pins keep the id
+      // (blockrun redirects them server-side to gpt-oss-120b).
       "nvidia/qwen3-next-80b-a3b-thinking": "free/qwen3-next-80b-a3b-instruct",
-      // New live free models from the 2026-06-14 BlockRun free-tier refresh
       "nvidia/qwen3-next-80b-a3b-instruct": "free/qwen3-next-80b-a3b-instruct",
       "nvidia/seed-oss-36b": "free/seed-oss-36b",
       "nvidia/mistral-nemotron": "free/mistral-nemotron",
       "nvidia/step-3.7-flash": "free/step-3.7-flash",
       "nvidia/nemotron-nano-9b-v2": "free/nemotron-nano-9b-v2",
       "nvidia/nemotron-nano-12b-v2-vl": "free/nemotron-nano-12b-v2-vl",
-      "nvidia/mistral-small-4-119b": "free/llama-4-maverick",
-      // NVIDIA upstream timing out — hidden + redirected server-side 2026-06-08
-      // Retired free IDs → successors (mirror server-side redirects)
-      "nvidia/nemotron-ultra-253b": "free/llama-4-maverick",
-      "nvidia/nemotron-3-super-120b": "free/llama-4-maverick",
-      "nvidia/nemotron-super-49b": "free/llama-4-maverick",
+      // nvidia/mistral-small-4-119b: no local redirect — recovered in the 2026-07-17
+      // re-probe and blockrun removed its server redirect, so pinned callers reach the
+      // real model again (upstream fallbackModel + health gate cover a relapse).
+      // Retired free IDs → successors (mirror blockrun's 2026-07-17 redirect map:
+      // ultra-253b redirects to gpt-oss-120b; the two supers have NO server redirect —
+      // they stay hidden-but-routable, so their pins pass through to the real ids).
+      "nvidia/nemotron-ultra-253b": "free/gpt-oss-120b",
       // mistral-large-3-675b un-retired 2026-06-14: BlockRun re-featured it (available,
       // NVIDIA upstream recovered) so it's a real free catalog entry again.
       "nvidia/mistral-large-3-675b": "free/mistral-large-3-675b",
       "nvidia/qwen3.5-122b-a10b": "free/qwen3.5-122b-a10b",
-      "nvidia/devstral-2-123b": "free/seed-oss-36b",
-      "free/nemotron-ultra-253b": "free/llama-4-maverick",
-      "free/nemotron-3-super-120b": "free/llama-4-maverick",
-      "free/nemotron-super-49b": "free/llama-4-maverick",
-      "free/devstral-2-123b": "free/seed-oss-36b",
+      // devstral-2-123b died upstream; blockrun redirects it to gpt-oss-120b (2026-07-17 map)
+      "nvidia/devstral-2-123b": "free/gpt-oss-120b",
+      "free/nemotron-ultra-253b": "free/gpt-oss-120b",
+      "free/devstral-2-123b": "free/gpt-oss-120b",
       // New blockrun-featured free models (2026-06-14 catalog sweep)
       "mistral-large": "free/mistral-large-3-675b",
       "mistral-large-3-675b": "free/mistral-large-3-675b",
       "qwen3.5-122b": "free/qwen3.5-122b-a10b",
       "qwen3-122b": "free/qwen3.5-122b-a10b",
-      // Free model shorthand aliases
+      // Free model shorthand aliases (v4-flash recovered 2026-07-17 — live again)
       "deepseek-free": "free/deepseek-v4-flash",
-      // V4 Pro NVIDIA hung 2026-04-30 → flash
       "deepseek-v4-pro": "free/deepseek-v4-flash",
-      // free shorthand → flash (pro hung)
+      // free shorthand → flash (pro still hung)
       "deepseek-v4-flash": "free/deepseek-v4-flash",
       "v4-pro": "free/deepseek-v4-flash",
       // V4 Pro NVIDIA hung → flash
       "v4-flash": "free/deepseek-v4-flash",
-      "mistral-free": "free/llama-4-maverick",
+      "mistral-free": "free/mistral-large-3-675b",
+      // was llama-4-maverick (died 2026-07) — live Mistral flagship
       "glm-free": "free/seed-oss-36b",
       // qwen3-coder retired → live coder successor
-      "llama-free": "free/llama-4-maverick",
+      "llama-free": "free/gpt-oss-120b",
+      // no live free Llama left (maverick died 2026-07)
       "qwen-coder": "free/seed-oss-36b",
       // qwen3-coder-480b retired → seed-oss-36b
       "qwen-coder-free": "free/seed-oss-36b",
-      "qwen-thinking": "free/qwen3-next-80b-a3b-instruct",
+      "qwen-thinking": "free/gpt-oss-120b",
+      // qwen3-next died 2026-07-17; no live free Qwen left
       "qwen3-next": "free/qwen3-next-80b-a3b-instruct",
-      // now a live model (instruct)
+      // explicit-ish pin — gateway redirects
       "qwen3-next-80b": "free/qwen3-next-80b-a3b-instruct",
-      "mistral-small": "free/llama-4-maverick",
+      "mistral-small": "free/mistral-nemotron",
+      // closest live Mistral-family free model
       // New live free models (2026-06-14 BlockRun free-tier refresh)
       "seed-oss": "free/seed-oss-36b",
       "seed-oss-36b": "free/seed-oss-36b",
@@ -34652,16 +34657,19 @@ var init_models = __esm({
       "nemotron-omni": "free/nemotron-3-nano-omni-30b-a3b-reasoning",
       "nano-omni": "free/nemotron-3-nano-omni-30b-a3b-reasoning",
       "vision-free": "free/nemotron-3-nano-omni-30b-a3b-reasoning",
-      // Retired shorthand aliases redirect to successors
-      nemotron: "free/llama-4-maverick",
-      "nemotron-ultra": "free/llama-4-maverick",
-      "nemotron-253b": "free/llama-4-maverick",
-      "nemotron-super": "free/llama-4-maverick",
-      "nemotron-49b": "free/llama-4-maverick",
-      "nemotron-120b": "free/llama-4-maverick",
+      // Retired shorthand aliases redirect to live successors (2026-07-17 map:
+      // llama-4-maverick died, so the old catch-all target moved to gpt-oss-120b)
+      nemotron: "free/nemotron-3-nano-omni-30b-a3b-reasoning",
+      // strongest live Nemotron
+      "nemotron-ultra": "free/gpt-oss-120b",
+      "nemotron-253b": "free/gpt-oss-120b",
+      "nemotron-super": "free/gpt-oss-120b",
+      "nemotron-49b": "free/gpt-oss-120b",
+      "nemotron-120b": "free/gpt-oss-120b",
       devstral: "free/seed-oss-36b",
       "devstral-2": "free/seed-oss-36b",
       maverick: "free/llama-4-maverick",
+      // explicit-ish pin — gateway redirects
       free: "free/gpt-oss-120b",
       // MiniMax (minimax → current flagship: M3)
       minimax: "minimax/minimax-m3",
@@ -35287,10 +35295,29 @@ var init_models = __esm({
         agentic: true,
         toolCalling: true
       },
-      // Kimi K2.7 — Moonshot's flagship (added 2026-06-13). 256K context, multi-modal
-      // (image + VIDEO input), returns reasoning_content. Served via BlockRun's OpenRouter
-      // credit pool (slug moonshotai/kimi-k2.7-code) failing over to direct Moonshot.
-      // AT-COST pricing ($0.95/$4.00 = OpenRouter COGS, zero margin) — same as K2.6.
+      // Kimi K3 — Moonshot's flagship (blockrun added 2026-07-17, live-probed same day).
+      // 2.8T-param open MoE, 1M context, image + text input, returns reasoning_content.
+      // Priced at COGS + BlockRun's 5% margin: users pay ~$3.15/$15.75 per 1M; the fields
+      // here store the raw $3.00/$15.00 COST (server applies margin at billing). ~5x K2.7,
+      // so the generic `kimi` alias deliberately stays on K2.7 — address k3 explicitly.
+      {
+        id: "moonshot/kimi-k3",
+        name: "Kimi K3",
+        version: "k3",
+        inputPrice: 3,
+        outputPrice: 15,
+        contextWindow: 1048576,
+        maxOutput: 65536,
+        reasoning: true,
+        vision: true,
+        agentic: true,
+        toolCalling: true
+      },
+      // Kimi K2.7 — previous-gen flagship (added 2026-06-13); superseded by K3 + hidden on
+      // BlockRun 2026-07-17 but kept routable. 256K context, multi-modal (image + VIDEO
+      // input), returns reasoning_content. Served via BlockRun's OpenRouter credit pool
+      // (slug moonshotai/kimi-k2.7-code) failing over to direct Moonshot. AT-COST pricing
+      // ($0.95/$4.00 = OpenRouter COGS, zero margin) — same as K2.6.
       {
         id: "moonshot/kimi-k2.7",
         name: "Kimi K2.7",
@@ -35590,6 +35617,8 @@ var init_models = __esm({
         // V4 Flash: 284B / 13B active MoE, 1M context. ~5x faster than V4 Pro.
         // Strong on chat/summarization (MMLU-Pro 86.2). Caveat: weaker factual
         // recall (SimpleQA 34% vs Pro's 58%) — pick V4 Pro for fact-heavy loops.
+        // Recovered in blockrun's 2026-07-17 live re-probe (3.2s both passes) —
+        // un-hidden upstream and back in the advertised free 8.
         id: "free/deepseek-v4-flash",
         name: "[Free] DeepSeek V4 Flash",
         version: "v4-flash",
@@ -35670,10 +35699,12 @@ var init_models = __esm({
       },
       // 2026-06-16: BlockRun's 2026-06-14 free-tier refresh (self-healing health gate
       // + probe-verified lineup, blockrun commit 5817ecd) added these live NVIDIA free
-      // models. qwen3-coder-480b was retired (NVIDIA EOL 2026-06-14) and now redirects
-      // server-side to seed-oss-36b; deepseek-v4-flash/glm-4.7 stay catalog-only.
+      // models. Status per blockrun's 2026-07-17 live re-probe: qwen3-coder-480b and
+      // glm-4.7 stay dead (server-redirected); deepseek-v4-flash recovered (see above).
       {
-        // Qwen3-Next 80B (A3B active MoE): 262K context, strong reasoning + coding.
+        // Qwen3-Next 80B (A3B active MoE): 262K context. DIED in the 2026-07-17
+        // re-probe (">60s / DEGRADED") — hidden upstream, gateway redirects pinned
+        // callers to gpt-oss-120b. Entry kept so pins stay routable; off the picker.
         id: "free/qwen3-next-80b-a3b-instruct",
         name: "[Free] Qwen3-Next 80B Instruct",
         version: "next-80b-a3b",
@@ -84552,6 +84583,7 @@ function estimateAmount(modelId, bodyLength, maxTokens) {
     const estimatedOutputTokens = maxTokens || model.maxOutput || 4096;
     costUsd = estimatedInputTokens / 1e6 * model.inputPrice + estimatedOutputTokens / 1e6 * model.outputPrice;
   }
+  if (costUsd > 0) costUsd += 2e-3;
   const amountMicros = Math.max(1e3, Math.ceil(costUsd * 1.2 * 1e6));
   return amountMicros.toString();
 }
@@ -88009,12 +88041,8 @@ var init_proxy = __esm({
       "free/gpt-oss-20b",
       "free/mistral-large-3-675b",
       // 675B general flagship (re-featured 2026-06-14)
-      "free/qwen3.5-122b-a10b",
-      // newest-gen Qwen, strong general
-      "free/qwen3-next-80b-a3b-instruct",
-      // 262K ctx, strong reasoning + coding
-      "free/llama-4-maverick",
-      // BlockRun's primary free fallback
+      "free/deepseek-v4-flash",
+      // 1M ctx, recovered 2026-07-17 (3.2s probe)
       "free/seed-oss-36b",
       // live coder (successor to retired qwen3-coder-480b)
       "free/mistral-nemotron",

package/dist/index.js CHANGED Viewed

@@ -85,12 +85,12 @@ var init_top_models = __esm({
       "zai/glm-5",
       "minimax/minimax-m3",
       "minimax/minimax-m2.7",
-      "moonshot/kimi-k2.7",
+      "moonshot/kimi-k3",
       "deepseek/deepseek-v4-pro",
       "deepseek/deepseek-chat",
       "deepseek/deepseek-reasoner",
       "free/mistral-large-3-675b",
-      "free/qwen3-next-80b-a3b-instruct",
+      "free/deepseek-v4-flash",
       "free/seed-oss-36b",
       "free/nemotron-3-nano-omni-30b-a3b-reasoning",
       "free/mistral-nemotron",
@@ -297,13 +297,16 @@ var init_models = __esm({
       deepseek: "deepseek/deepseek-chat",
       "deepseek-chat": "deepseek/deepseek-chat",
       reasoner: "deepseek/deepseek-reasoner",
-      // Kimi / Moonshot — K2.7 is the featured flagship on BlockRun (added 2026-06-13,
-      // commit cd3d79b; K2.6 marked hidden/superseded, K2.5 hidden). Bare aliases now
-      // resolve to K2.7 (same $0.95/$4.00 price as K2.6 — AT-COST, zero margin). Explicit
+      // Kimi / Moonshot — K3 is the featured flagship on BlockRun (added 2026-07-17; K2.7
+      // hidden/superseded, K2.6 hidden, K2.5 hidden). K3 is ~5x K2.7's price ($3/$15 vs
+      // $0.95/$4.00), so the BARE aliases deliberately STAY on K2.7 — repointing them to K3
+      // would silently ~5x every generic-`kimi` quote and break per-call-cap wallets (mirrors
+      // blockrun's own alias decision). Address the flagship explicitly via "kimi-k3". Explicit
       // pins for "kimi-k2.6" / "kimi-k2.5" still resolve to those exact models (K2.5 is a
       // cost-stability opt-in at $0.60/$3.00). NVIDIA-hosted K2.5 was retired 2026-04-21.
       kimi: "moonshot/kimi-k2.7",
       moonshot: "moonshot/kimi-k2.7",
+      "kimi-k3": "moonshot/kimi-k3",
       "kimi-k2": "moonshot/kimi-k2.7",
       "kimi-k2.7": "moonshot/kimi-k2.7",
       "kimi-k2.6": "moonshot/kimi-k2.6",
@@ -360,14 +363,12 @@ var init_models = __esm({
       "gpt-20b": "free/gpt-oss-20b",
       "nvidia/gpt-oss-120b": "free/gpt-oss-120b",
       "nvidia/gpt-oss-20b": "free/gpt-oss-20b",
+      // deepseek family → v4-flash, matching blockrun's 2026-07-17 redirect map
+      // (v4-flash recovered in the live re-probe; v3.2/v4-pro stay dead upstream).
       "nvidia/deepseek-v3.2": "free/deepseek-v4-flash",
-      // v3.2 → v4-flash (v4-pro NVIDIA hung 2026-04-30)
       "free/deepseek-v3.2": "free/deepseek-v4-flash",
-      // local pin redirect
       "nvidia/deepseek-v4-pro": "free/deepseek-v4-flash",
-      // V4 Pro NVIDIA hung 2026-04-30
       "free/deepseek-v4-pro": "free/deepseek-v4-flash",
-      // V4 Pro delisted — redirect pinned callers
       "nvidia/deepseek-v4-flash": "free/deepseek-v4-flash",
       "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning": "free/nemotron-3-nano-omni-30b-a3b-reasoning",
       // qwen3-coder-480b retired (NVIDIA EOL 2026-06-14) → redirects to seed-oss-36b
@@ -375,58 +376,62 @@ var init_models = __esm({
       // resolves them), but point the generic coding shorthands at the live successor.
       "nvidia/qwen3-coder-480b": "free/qwen3-coder-480b",
       "qwen/qwen3-coder-480b-a35b-instruct": "free/qwen3-coder-480b",
-      "nvidia/glm-4.7": "free/seed-oss-36b",
+      // glm-4.7 pin keeps the real id — blockrun redirects it server-side (→ gpt-oss-120b).
+      "nvidia/glm-4.7": "free/glm-4.7",
       "nvidia/llama-4-maverick": "free/llama-4-maverick",
-      // qwen3-next thinking variant retired; live sibling is the instruct model
+      // qwen3-next: both variants died in the 2026-07-17 re-probe; pins keep the id
+      // (blockrun redirects them server-side to gpt-oss-120b).
       "nvidia/qwen3-next-80b-a3b-thinking": "free/qwen3-next-80b-a3b-instruct",
-      // New live free models from the 2026-06-14 BlockRun free-tier refresh
       "nvidia/qwen3-next-80b-a3b-instruct": "free/qwen3-next-80b-a3b-instruct",
       "nvidia/seed-oss-36b": "free/seed-oss-36b",
       "nvidia/mistral-nemotron": "free/mistral-nemotron",
       "nvidia/step-3.7-flash": "free/step-3.7-flash",
       "nvidia/nemotron-nano-9b-v2": "free/nemotron-nano-9b-v2",
       "nvidia/nemotron-nano-12b-v2-vl": "free/nemotron-nano-12b-v2-vl",
-      "nvidia/mistral-small-4-119b": "free/llama-4-maverick",
-      // NVIDIA upstream timing out — hidden + redirected server-side 2026-06-08
-      // Retired free IDs → successors (mirror server-side redirects)
-      "nvidia/nemotron-ultra-253b": "free/llama-4-maverick",
-      "nvidia/nemotron-3-super-120b": "free/llama-4-maverick",
-      "nvidia/nemotron-super-49b": "free/llama-4-maverick",
+      // nvidia/mistral-small-4-119b: no local redirect — recovered in the 2026-07-17
+      // re-probe and blockrun removed its server redirect, so pinned callers reach the
+      // real model again (upstream fallbackModel + health gate cover a relapse).
+      // Retired free IDs → successors (mirror blockrun's 2026-07-17 redirect map:
+      // ultra-253b redirects to gpt-oss-120b; the two supers have NO server redirect —
+      // they stay hidden-but-routable, so their pins pass through to the real ids).
+      "nvidia/nemotron-ultra-253b": "free/gpt-oss-120b",
       // mistral-large-3-675b un-retired 2026-06-14: BlockRun re-featured it (available,
       // NVIDIA upstream recovered) so it's a real free catalog entry again.
       "nvidia/mistral-large-3-675b": "free/mistral-large-3-675b",
       "nvidia/qwen3.5-122b-a10b": "free/qwen3.5-122b-a10b",
-      "nvidia/devstral-2-123b": "free/seed-oss-36b",
-      "free/nemotron-ultra-253b": "free/llama-4-maverick",
-      "free/nemotron-3-super-120b": "free/llama-4-maverick",
-      "free/nemotron-super-49b": "free/llama-4-maverick",
-      "free/devstral-2-123b": "free/seed-oss-36b",
+      // devstral-2-123b died upstream; blockrun redirects it to gpt-oss-120b (2026-07-17 map)
+      "nvidia/devstral-2-123b": "free/gpt-oss-120b",
+      "free/nemotron-ultra-253b": "free/gpt-oss-120b",
+      "free/devstral-2-123b": "free/gpt-oss-120b",
       // New blockrun-featured free models (2026-06-14 catalog sweep)
       "mistral-large": "free/mistral-large-3-675b",
       "mistral-large-3-675b": "free/mistral-large-3-675b",
       "qwen3.5-122b": "free/qwen3.5-122b-a10b",
       "qwen3-122b": "free/qwen3.5-122b-a10b",
-      // Free model shorthand aliases
+      // Free model shorthand aliases (v4-flash recovered 2026-07-17 — live again)
       "deepseek-free": "free/deepseek-v4-flash",
-      // V4 Pro NVIDIA hung 2026-04-30 → flash
       "deepseek-v4-pro": "free/deepseek-v4-flash",
-      // free shorthand → flash (pro hung)
+      // free shorthand → flash (pro still hung)
       "deepseek-v4-flash": "free/deepseek-v4-flash",
       "v4-pro": "free/deepseek-v4-flash",
       // V4 Pro NVIDIA hung → flash
       "v4-flash": "free/deepseek-v4-flash",
-      "mistral-free": "free/llama-4-maverick",
+      "mistral-free": "free/mistral-large-3-675b",
+      // was llama-4-maverick (died 2026-07) — live Mistral flagship
       "glm-free": "free/seed-oss-36b",
       // qwen3-coder retired → live coder successor
-      "llama-free": "free/llama-4-maverick",
+      "llama-free": "free/gpt-oss-120b",
+      // no live free Llama left (maverick died 2026-07)
       "qwen-coder": "free/seed-oss-36b",
       // qwen3-coder-480b retired → seed-oss-36b
       "qwen-coder-free": "free/seed-oss-36b",
-      "qwen-thinking": "free/qwen3-next-80b-a3b-instruct",
+      "qwen-thinking": "free/gpt-oss-120b",
+      // qwen3-next died 2026-07-17; no live free Qwen left
       "qwen3-next": "free/qwen3-next-80b-a3b-instruct",
-      // now a live model (instruct)
+      // explicit-ish pin — gateway redirects
       "qwen3-next-80b": "free/qwen3-next-80b-a3b-instruct",
-      "mistral-small": "free/llama-4-maverick",
+      "mistral-small": "free/mistral-nemotron",
+      // closest live Mistral-family free model
       // New live free models (2026-06-14 BlockRun free-tier refresh)
       "seed-oss": "free/seed-oss-36b",
       "seed-oss-36b": "free/seed-oss-36b",
@@ -442,16 +447,19 @@ var init_models = __esm({
       "nemotron-omni": "free/nemotron-3-nano-omni-30b-a3b-reasoning",
       "nano-omni": "free/nemotron-3-nano-omni-30b-a3b-reasoning",
       "vision-free": "free/nemotron-3-nano-omni-30b-a3b-reasoning",
-      // Retired shorthand aliases redirect to successors
-      nemotron: "free/llama-4-maverick",
-      "nemotron-ultra": "free/llama-4-maverick",
-      "nemotron-253b": "free/llama-4-maverick",
-      "nemotron-super": "free/llama-4-maverick",
-      "nemotron-49b": "free/llama-4-maverick",
-      "nemotron-120b": "free/llama-4-maverick",
+      // Retired shorthand aliases redirect to live successors (2026-07-17 map:
+      // llama-4-maverick died, so the old catch-all target moved to gpt-oss-120b)
+      nemotron: "free/nemotron-3-nano-omni-30b-a3b-reasoning",
+      // strongest live Nemotron
+      "nemotron-ultra": "free/gpt-oss-120b",
+      "nemotron-253b": "free/gpt-oss-120b",
+      "nemotron-super": "free/gpt-oss-120b",
+      "nemotron-49b": "free/gpt-oss-120b",
+      "nemotron-120b": "free/gpt-oss-120b",
       devstral: "free/seed-oss-36b",
       "devstral-2": "free/seed-oss-36b",
       maverick: "free/llama-4-maverick",
+      // explicit-ish pin — gateway redirects
       free: "free/gpt-oss-120b",
       // MiniMax (minimax → current flagship: M3)
       minimax: "minimax/minimax-m3",
@@ -1077,10 +1085,29 @@ var init_models = __esm({
         agentic: true,
         toolCalling: true
       },
-      // Kimi K2.7 — Moonshot's flagship (added 2026-06-13). 256K context, multi-modal
-      // (image + VIDEO input), returns reasoning_content. Served via BlockRun's OpenRouter
-      // credit pool (slug moonshotai/kimi-k2.7-code) failing over to direct Moonshot.
-      // AT-COST pricing ($0.95/$4.00 = OpenRouter COGS, zero margin) — same as K2.6.
+      // Kimi K3 — Moonshot's flagship (blockrun added 2026-07-17, live-probed same day).
+      // 2.8T-param open MoE, 1M context, image + text input, returns reasoning_content.
+      // Priced at COGS + BlockRun's 5% margin: users pay ~$3.15/$15.75 per 1M; the fields
+      // here store the raw $3.00/$15.00 COST (server applies margin at billing). ~5x K2.7,
+      // so the generic `kimi` alias deliberately stays on K2.7 — address k3 explicitly.
+      {
+        id: "moonshot/kimi-k3",
+        name: "Kimi K3",
+        version: "k3",
+        inputPrice: 3,
+        outputPrice: 15,
+        contextWindow: 1048576,
+        maxOutput: 65536,
+        reasoning: true,
+        vision: true,
+        agentic: true,
+        toolCalling: true
+      },
+      // Kimi K2.7 — previous-gen flagship (added 2026-06-13); superseded by K3 + hidden on
+      // BlockRun 2026-07-17 but kept routable. 256K context, multi-modal (image + VIDEO
+      // input), returns reasoning_content. Served via BlockRun's OpenRouter credit pool
+      // (slug moonshotai/kimi-k2.7-code) failing over to direct Moonshot. AT-COST pricing
+      // ($0.95/$4.00 = OpenRouter COGS, zero margin) — same as K2.6.
       {
         id: "moonshot/kimi-k2.7",
         name: "Kimi K2.7",
@@ -1380,6 +1407,8 @@ var init_models = __esm({
         // V4 Flash: 284B / 13B active MoE, 1M context. ~5x faster than V4 Pro.
         // Strong on chat/summarization (MMLU-Pro 86.2). Caveat: weaker factual
         // recall (SimpleQA 34% vs Pro's 58%) — pick V4 Pro for fact-heavy loops.
+        // Recovered in blockrun's 2026-07-17 live re-probe (3.2s both passes) —
+        // un-hidden upstream and back in the advertised free 8.
         id: "free/deepseek-v4-flash",
         name: "[Free] DeepSeek V4 Flash",
         version: "v4-flash",
@@ -1460,10 +1489,12 @@ var init_models = __esm({
       },
       // 2026-06-16: BlockRun's 2026-06-14 free-tier refresh (self-healing health gate
       // + probe-verified lineup, blockrun commit 5817ecd) added these live NVIDIA free
-      // models. qwen3-coder-480b was retired (NVIDIA EOL 2026-06-14) and now redirects
-      // server-side to seed-oss-36b; deepseek-v4-flash/glm-4.7 stay catalog-only.
+      // models. Status per blockrun's 2026-07-17 live re-probe: qwen3-coder-480b and
+      // glm-4.7 stay dead (server-redirected); deepseek-v4-flash recovered (see above).
       {
-        // Qwen3-Next 80B (A3B active MoE): 262K context, strong reasoning + coding.
+        // Qwen3-Next 80B (A3B active MoE): 262K context. DIED in the 2026-07-17
+        // re-probe (">60s / DEGRADED") — hidden upstream, gateway redirects pinned
+        // callers to gpt-oss-120b. Entry kept so pins stay routable; off the picker.
         id: "free/qwen3-next-80b-a3b-instruct",
         name: "[Free] Qwen3-Next 80B Instruct",
         version: "next-80b-a3b",
@@ -35452,10 +35483,10 @@ var init_config = __esm({
             // FREE — smaller, faster
             "free/mistral-large-3-675b",
             // FREE — 675B general flagship (re-featured 2026-06-14)
-            "free/qwen3.5-122b-a10b",
-            // FREE — newest-gen Qwen, strong general
-            "free/qwen3-next-80b-a3b-instruct",
-            // FREE — 262K ctx, strong general (deepseek-v4-flash retired)
+            "free/deepseek-v4-flash",
+            // FREE — 1M ctx, recovered in blockrun's 2026-07-17 re-probe
+            "free/seed-oss-36b",
+            // FREE — live coder (qwen3.5-122b + qwen3-next died 2026-07)
             "google/gemini-3.1-flash-lite",
             // $0.25/$1.50 — newest flash-lite
             "openai/gpt-5.4-nano",
@@ -84372,6 +84403,7 @@ function estimateAmount(modelId, bodyLength, maxTokens) {
     const estimatedOutputTokens = maxTokens || model.maxOutput || 4096;
     costUsd = estimatedInputTokens / 1e6 * model.inputPrice + estimatedOutputTokens / 1e6 * model.outputPrice;
   }
+  if (costUsd > 0) costUsd += 2e-3;
   const amountMicros = Math.max(1e3, Math.ceil(costUsd * 1.2 * 1e6));
   return amountMicros.toString();
 }
@@ -87829,12 +87861,8 @@ var init_proxy = __esm({
       "free/gpt-oss-20b",
       "free/mistral-large-3-675b",
       // 675B general flagship (re-featured 2026-06-14)
-      "free/qwen3.5-122b-a10b",
-      // newest-gen Qwen, strong general
-      "free/qwen3-next-80b-a3b-instruct",
-      // 262K ctx, strong reasoning + coding
-      "free/llama-4-maverick",
-      // BlockRun's primary free fallback
+      "free/deepseek-v4-flash",
+      // 1M ctx, recovered 2026-07-17 (3.2s probe)
       "free/seed-oss-36b",
       // live coder (successor to retired qwen3-coder-480b)
       "free/mistral-nemotron",

package/package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "@blockrun/clawrouter",
-  "version": "0.12.227",
+  "version": "0.12.229",
   "description": "Smart LLM router — save 85% on inference costs. 55+ models (8 free), one wallet, x402 micropayments.",
   "type": "module",
   "main": "dist/index.js",

package/skills/clawrouter/SKILL.md CHANGED Viewed

@@ -135,7 +135,7 @@ Rules handle ~~80% of requests in <1ms. Only ambiguous queries hit the LLM class
 ## Available Models
-55+ models including: gpt-5.6-terra [balanced, stable default], gpt-5.6-sol [flagship], gpt-5.6-luna [cost-efficient], gpt-5.5, gpt-5.4, gpt-4o, o3, claude-fable-5, claude-opus-4.8, claude-opus-4.7, claude-opus-4.5, claude-sonnet-5, claude-sonnet-4.6, gemini-3.1-pro, gemini-2.5-flash, deepseek-v4-pro, deepseek-chat, grok-4.5, grok-4.3, grok-build-0.1, kimi-k2.7, and 8 free NVIDIA models (mistral-large-3-675b [default, 675B flagship], qwen3-next-80b-a3b-instruct, seed-oss-36b [coding], nemotron-3-nano-omni-30b-a3b-reasoning [vision], mistral-nemotron, step-3.7-flash, nemotron-nano-9b-v2, nemotron-nano-12b-v2-vl [vision]).
+55+ models including: gpt-5.6-terra [balanced, stable default], gpt-5.6-sol [flagship], gpt-5.6-luna [cost-efficient], gpt-5.5, gpt-5.4, gpt-4o, o3, claude-fable-5, claude-opus-4.8, claude-opus-4.7, claude-opus-4.5, claude-sonnet-5, claude-sonnet-4.6, gemini-3.1-pro, gemini-2.5-flash, deepseek-v4-pro, deepseek-chat, grok-4.5, grok-4.3, grok-build-0.1, kimi-k3 [1M ctx flagship], kimi-k2.7, and 8 free NVIDIA models (mistral-large-3-675b [default, 675B flagship], deepseek-v4-flash [1M context], seed-oss-36b [coding], nemotron-3-nano-omni-30b-a3b-reasoning [vision], mistral-nemotron, step-3.7-flash, nemotron-nano-9b-v2, nemotron-nano-12b-v2-vl [vision]).
 ## Built-in Agent Tools
@@ -158,7 +158,7 @@ Realtime prices and historical OHLC across every asset class. The agent should c
 | Tool                        | Purpose                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                      | Price                                                               |
 | --------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------- |
-| `blockrun_image_generation` | 8 image models — GPT Image 1/2, Nano Banana / Pro, Seedream 5 Pro, Grok Imagine / Pro, CogView-4                                                                                                                                                                                                                                                                                                                                                                                                                                                                                | $0.015–$0.15 / image                                                |
+| `blockrun_image_generation` | 8 image models — GPT Image 1/2, Nano Banana / Pro, Seedream 5 Pro, Grok Imagine / Pro, CogView-4                                                                                                                                                                                                                                                                                                                                                                                                                                                             | $0.015–$0.15 / image                                                |
 | `blockrun_image_edit`       | Edit / inpaint existing image (openai/gpt-image-1)                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                           | $0.02–$0.04 / image                                                 |
 | `blockrun_video_generation` | Grok Imagine + ByteDance Seedance (1.5-pro / 2.0-fast / 2.0) at 720p with synced audio (t2v default), 5–10s, + OpenAI Sora 2 via Azure (`azure/sora-2`, 4/8/12s, t2v + i2v, rejects human faces in reference images). Seedance is token-priced upstream (~20,256 tokens/sec at the 720p+audio default — 2× the prior 480p rate); `image_url` (image-to-video) costs the same as t2v since 2026-06-01. Seedance 2.0 variants accept optional `real_face_asset_id` (`ta_…`) for BytePlus RealFace character-consistency — mutually exclusive with `image_url`. | $0.05/s (Grok); $0.10/s (Sora 2); Seedance ~$0.46–$1.49 per 5s clip |

package/skills/imagegen/SKILL.md CHANGED Viewed

@@ -55,16 +55,16 @@ Display inline: `![generated image](http://localhost:8402/images/abc123.png)`
 ### Model Selection
-| Alias              | Full ID                      | Price        | Sizes                           | Best for                                                         |
-| ------------------ | ---------------------------- | ------------ | ------------------------------- | ---------------------------------------------------------------- |
-| `nano-banana`      | `google/nano-banana`         | $0.05        | 1024×1024, 1216×832, 1024×1792  | Default — fast, cheap, good quality                              |
-| `banana-pro`       | `google/nano-banana-pro`     | $0.10–$0.15  | up to 4096×4096                 | High-res, large format                                           |
-| `gpt-image`        | `openai/gpt-image-1`         | $0.02–$0.04  | 1024×1024, 1536×1024, 1024×1536 | Budget option; supports editing                                  |
+| Alias              | Full ID                      | Price        | Sizes                           | Best for                                                                                                           |
+| ------------------ | ---------------------------- | ------------ | ------------------------------- | ------------------------------------------------------------------------------------------------------------------ |
+| `nano-banana`      | `google/nano-banana`         | $0.05        | 1024×1024, 1216×832, 1024×1792  | Default — fast, cheap, good quality                                                                                |
+| `banana-pro`       | `google/nano-banana-pro`     | $0.10–$0.15  | up to 4096×4096                 | High-res, large format                                                                                             |
+| `gpt-image`        | `openai/gpt-image-1`         | $0.02–$0.04  | 1024×1024, 1536×1024, 1024×1536 | Budget option; supports editing                                                                                    |
 | `gpt-image-2`      | `openai/gpt-image-2`         | $0.06–$0.12  | 1024×1024, 1536×1024, 1024×1536 | Photorealistic, reasoning-driven, text rendering (slow — proxy polls up to 5min); legacy `dalle` alias routes here |
-| `seedream`         | `bytedance/seedream-5-pro`   | $0.045–$0.09 | up to 2848×1600 / 2304×1728     | Flagship quality, reference-image support                        |
-| `grok-imagine`     | `xai/grok-imagine-image`     | $0.02        | 1024×1024                       | xAI Grok image style                                             |
-| `grok-imagine-pro` | `xai/grok-imagine-image-pro` | $0.07        | 1024×1024                       | Grok high-quality                                                |
-| `cogview`          | `zai/cogview-4`              | $0.015–$0.02 | 512×512 to 1440×1440            | Cheapest — Zhipu CogView                                         |
+| `seedream`         | `bytedance/seedream-5-pro`   | $0.045–$0.09 | up to 2848×1600 / 2304×1728     | Flagship quality, reference-image support                                                                          |
+| `grok-imagine`     | `xai/grok-imagine-image`     | $0.02        | 1024×1024                       | xAI Grok image style                                                                                               |
+| `grok-imagine-pro` | `xai/grok-imagine-image-pro` | $0.07        | 1024×1024                       | Grok high-quality                                                                                                  |
+| `cogview`          | `zai/cogview-4`              | $0.015–$0.02 | 512×512 to 1440×1440            | Cheapest — Zhipu CogView                                                                                           |
 **Choosing a model:**

package/src/top-models.json CHANGED Viewed

@@ -31,12 +31,12 @@
   "zai/glm-5",
   "minimax/minimax-m3",
   "minimax/minimax-m2.7",
-  "moonshot/kimi-k2.7",
+  "moonshot/kimi-k3",
   "deepseek/deepseek-v4-pro",
   "deepseek/deepseek-chat",
   "deepseek/deepseek-reasoner",
   "free/mistral-large-3-675b",
-  "free/qwen3-next-80b-a3b-instruct",
+  "free/deepseek-v4-flash",
   "free/seed-oss-36b",
   "free/nemotron-3-nano-omni-30b-a3b-reasoning",
   "free/mistral-nemotron",