pi-fireworks-provider 1.6.2 → 1.7.0

package/README.md

@@ -15,10 +15,10 @@ _Kimi, MiniMax, GLM, DeepSeek, GPT-OSS — via Fireworks AI's Anthropic Messages
 
 ## Features
 
-- **50+ AI Models** including Kimi K2.5, MiniMax M2.5, GLM 4.5/4.7/5, DeepSeek V3.1/V3.2, DeepSeek V4 Flash, and GPT-OSS
+- **52+ AI Models** including Kimi K2.5, MiniMax M2.5, GLM 4.5/4.7/5, DeepSeek V3.1/V3.2, DeepSeek V4 Flash, and GPT-OSS
 - **Dual API support** via Fireworks AI's Anthropic Messages and OpenAI-compatible completions endpoints (per-model routing, matching pi core's Fireworks provider)
 - **Service tiers** — toggle Fireworks `priority` vs `standard` per request on supported models (with priority pricing reflected in cost tracking), via a keybinding, `/fireworks-tier`, and a footer status area
-- **Preserved thinking** — toggle Fireworks' `reasoning_history: "preserved"` so prior assistant reasoning is retained across turns (better multi-turn recall; uses more tokens), via the `/fireworks-settings` panel, with a model-select notification. Matches neuralwatt/makora's settings-only UX, adapted to Fireworks' single global `reasoning_history` knob
+- **Preserved thinking** — toggle Fireworks' `reasoning_history: "preserved"` so prior assistant reasoning is retained across turns (better multi-turn recall; uses more tokens), via the `/fireworks-settings` panel, with a model-select notification. Applied only on models whose documented support includes `"preserved"` (Kimi K2.6/K2.7, GLM 5.2/4.7, …); interleaved-only models like DeepSeek V4 and MiniMax M2 keep their default. Matches neuralwatt/makora's settings-only UX, adapted to Fireworks' global `reasoning_history` knob
 - **Logit bias** — set an OpenAI-style `logit_bias` map (token ID → -100..100) via a nested `/fireworks-settings` panel (add / edit / delete arbitrary token IDs), sent on every Fireworks OpenAI-completions request; persisted to `~/.pi/agent/extensions/fireworks.json`
 - **Settings panel** — `/fireworks-settings` (TUI) to configure preserved thinking, logit bias, service tier, and display preferences; persisted to `~/.pi/agent/extensions/fireworks.json`
 - **Cost Tracking** with per-model pricing for budget management
@@ -75,11 +75,12 @@ pi
 | DeepSeek V3.1 | Text | 164K | 164K | $0.56 | $1.68 |
 | DeepSeek V3.2 | Text | 164K | 160K | $0.56 | $1.68 |
 | DeepSeek V4 Flash | Text | 1.0M | 384K | $0.14 | $0.28 |
-| DeepSeek V4 Flash 0731 | Text | 1.0M | 384K | $0.14 | $0.28 |
+| DeepSeek V4 Flash 0731 | Text | 1.0M | 384K | $0.22 | $0.66 |
+| DeepSeek V4 Flash Vision Exp | Text + Image | 1.0M | 384K | $0.22 | $0.66 |
 | DeepSeek V4 Pro | Text | 1.0M | 384K | $1.74 | $3.48 |
 | DeepSeek V4 Pro (router) | Text | 1.0M | 384K | $1.74 | $3.48 |
-| DeepSeek V4 Pro 0813 | Text | 1.0M | 384K | — | — |
-| DeepSeek-V4-Flash-Vision-Exp | Text + Image | 1.0M | 0 | — | — |
+| DeepSeek V4 Pro 0813 | Text | 1.0M | 384K | $1.32 | $3.96 |
+| DeepSeek V4.1 Flash | Text + Image | 1.0M | 384K | $0.22 | $0.66 |
 | Gemma 4 26B A4B IT | Text + Image | 262K | 33K | — | — |
 | Gemma 4 31B IT | Text + Image | 262K | 33K | — | — |
 | GLM 4.5 | Text | 131K | 131K | $0.55 | $2.19 |
@@ -114,13 +115,14 @@ pi
 | MiniMax-M2.5 | Text | 197K | 197K | $0.30 | $1.20 |
 | MiniMax-M2.7 | Text | 197K | 197K | $0.30 | $1.20 |
 | MiniMax-M3 | Text + Image | 512K | 512K | $0.30 | $1.20 |
-| Muse Glimmer 30B | Text + Image | 131K | 131K | — | — |
-| Nemotron 3.5 Lightning 30B A3B | Text | 262K | 262K | — | — |
+| Muse Glimmer 30B | Text + Image | 131K | 131K | $0.35 | $1.50 |
+| Nemotron 3.5 Lightning 30B A3B | Text | 262K | 262K | $0.05 | $0.20 |
 | NVIDIA Nemotron 3 Ultra NVFP4 | Text | 262K | 66K | $0.60 | $2.40 |
 | Qwen 3.7 Plus | Text + Image | 262K | 66K | $0.40 | $1.60 |
 | Qwen3 8B | Text | 41K | 41K | $0.20 | $0.20 |
 | Qwen3 VL 30B A3B Instruct | Text + Image | 262K | 33K | $0.50 | $0.50 |
 | Qwen3 VL 30B A3B Thinking | Text + Image | 262K | 33K | $0.50 | $0.50 |
+| Qwen3.8 Max | Text | 262K | 131K | $2.00 | $6.00 |
 | Qwen3.8-2.4T-A95B | Text | 262K | 131K | — | — |
 *Costs are per million tokens. Prices subject to change - check [fireworks.ai](https://fireworks.ai) for current pricing.*
 

package/custom-models.json

@@ -250,5 +250,21 @@
       "deferredToolsMode": "kimi",
       "supportsReasoningEffort": true
     }
+  },
+  {
+    "id": "accounts/fireworks/models/qwen3p8-max",
+    "name": "Qwen3.8 Max",
+    "reasoning": true,
+    "cost": {
+      "input": 2,
+      "output": 6,
+      "cacheRead": 0.25,
+      "cacheWrite": 0
+    },
+    "input": [
+      "text"
+    ],
+    "contextWindow": 262144,
+    "maxTokens": 131072
   }
 ]

package/index.ts

@@ -384,19 +384,28 @@ function stripAnchorBleedInPlace(obj: Record<string, unknown>): void {
 // orthogonal to the "fast" router model IDs (e.g. routers/...-fast), which are
 // separate models; priority applies to the base models below.
 //
-// Per-request priority pricing (USD per million tokens) from Fireworks' tier
-// reference. cacheWrite is not tiered (stays 0).
+// Per-model priority pricing (USD per million tokens, input / cached input /
+// output) from https://docs.fireworks.ai/serverless/pricing — that table is
+// the source of truth for the Priority serving path. Entries for the undated
+// DeepSeek aliases (not listed there) keep their previously curated values.
 const PRIORITY_PRICING: Record<string, { input: number; output: number; cacheRead: number; cacheWrite: number }> = {
-  "accounts/fireworks/models/glm-5p2":          { input: 1.75, output: 5.5,  cacheRead: 0.175, cacheWrite: 0 },
-  "accounts/fireworks/models/glm-5p3":          { input: 1.75, output: 5.5,  cacheRead: 0.325, cacheWrite: 0 },
-  "accounts/fireworks/models/kimi-k2p7-code":   { input: 1.43, output: 6,   cacheRead: 0.29,  cacheWrite: 0 },
-  "accounts/fireworks/models/minimax-m3":       { input: 0.45, output: 1.8,  cacheRead: 0.09,  cacheWrite: 0 },
-  "accounts/fireworks/models/deepseek-v4-pro":  { input: 2.61, output: 5.22, cacheRead: 0.218, cacheWrite: 0 },
-  "accounts/fireworks/models/kimi-k2p6":        { input: 1.5,  output: 6,   cacheRead: 0.22,  cacheWrite: 0 },
-  "accounts/fireworks/models/minimax-m2p7":    { input: 0.45, output: 1.8,  cacheRead: 0.09,  cacheWrite: 0 },
-  "accounts/fireworks/models/glm-5p1":         { input: 2.1,  output: 6.6,  cacheRead: 0.39,  cacheWrite: 0 },
-  "accounts/fireworks/models/gpt-oss-120b":     { input: 0.18, output: 0.72, cacheRead: 0.018, cacheWrite: 0 },
-  "accounts/fireworks/models/deepseek-v4-flash":{ input: 0.21, output: 0.42, cacheRead: 0.045, cacheWrite: 0 },
+  "accounts/fireworks/models/kimi-k3":                      { input: 3.75,  output: 18.75, cacheRead: 0.375,   cacheWrite: 0 },
+  "accounts/fireworks/models/kimi-k2p7-code":               { input: 1.425, output: 6,     cacheRead: 0.285,   cacheWrite: 0 },
+  "accounts/fireworks/models/kimi-k2p6":                    { input: 1.5,   output: 6,     cacheRead: 0.22,    cacheWrite: 0 },
+  "accounts/fireworks/models/deepseek-v4p1-flash":          { input: 0.275, output: 0.825, cacheRead: 0.00875, cacheWrite: 0 },
+  "accounts/fireworks/models/deepseek-v4-flash-0731":       { input: 0.275, output: 0.825, cacheRead: 0.00875, cacheWrite: 0 },
+  "accounts/fireworks/models/deepseek-v4-flash":            { input: 0.21,  output: 0.42,  cacheRead: 0.045,   cacheWrite: 0 },
+  "accounts/fireworks/models/deepseek-v4-flash-vision-exp": { input: 0.275, output: 0.825, cacheRead: 0.00875, cacheWrite: 0 },
+  "accounts/fireworks/models/deepseek-v4-pro-0813":         { input: 1.65,  output: 4.95,  cacheRead: 0.055,   cacheWrite: 0 },
+  "accounts/fireworks/models/deepseek-v4-pro":              { input: 2.61,  output: 5.22,  cacheRead: 0.218,   cacheWrite: 0 },
+  "accounts/fireworks/models/glm-5p2":                      { input: 1.75,  output: 5.5,   cacheRead: 0.18,    cacheWrite: 0 },
+  "accounts/fireworks/models/glm-5p3":                      { input: 1.75,  output: 5.5,   cacheRead: 0.325,   cacheWrite: 0 },
+  "accounts/fireworks/models/glm-5p1":                      { input: 2.1,   output: 6.6,   cacheRead: 0.39,    cacheWrite: 0 },
+  "accounts/fireworks/models/minimax-m3":                   { input: 0.45,  output: 1.8,   cacheRead: 0.09,    cacheWrite: 0 },
+  "accounts/fireworks/models/minimax-m2p7":                 { input: 0.45,  output: 1.8,   cacheRead: 0.09,    cacheWrite: 0 },
+  "accounts/fireworks/models/qwen3p8-max":                  { input: 3,     output: 9,     cacheRead: 0.375,   cacheWrite: 0 },
+  "accounts/fireworks/models/gpt-oss-120b":                 { input: 0.18,  output: 0.72,  cacheRead: 0.018,   cacheWrite: 0 },
+  "accounts/fireworks/models/muse-glimmer-30b":             { input: 0.525, output: 2.25,  cacheRead: 0.06,    cacheWrite: 0 },
 };
 
 type ServiceTier = "standard" | "priority";
@@ -698,8 +707,11 @@ function recomputePriorityCost(message: any): any {
 //
 // Unlike neuralwatt/makora (which use per-model vLLM `chat_template_kwargs`
 // flags like preserve_thinking/clear_thinking), Fireworks' knob is a single
-// global top-level param that applies to every reasoning model. We expose it
-// as one on/off toggle. See https://docs.fireworks.ai/guides/reasoning#preserved-thinking.
+// top-level param (`reasoning_history`). It accepts three values — "disabled",
+// "interleaved", and "preserved" — but per-model support varies (Fireworks API
+// reference). We expose the strongest mode as one on/off toggle and only send
+// "preserved" where the model accepts it. See
+// https://docs.fireworks.ai/guides/reasoning#preserved-thinking.
 
 // A Fireworks model is preserve-eligible if it's a reasoning model. We read
 // `reasoning` off the registered model when available, but also accept a
@@ -709,6 +721,30 @@ function isPreserveEligible(model: any): boolean {
   return model.reasoning === true;
 }
 
+// Models whose `reasoning_history` support does NOT include "preserved".
+// Fireworks currently documents MiniMax M2 and DeepSeek V4 as supporting only
+// "interleaved" (default) / "disabled"; Kimi K2.6/K2.7, Kimi K2 Instruct,
+// GLM 4.7/5.2, and Qwen 3.6 do support "preserved". Models outside the
+// documented table stay permissive (previous behavior) since the docs defer to
+// each provider's own documentation.
+const REASONING_HISTORY_NO_PRESERVE: ReadonlySet<string> = new Set([
+  "accounts/fireworks/models/minimax-m2p1",
+  "accounts/fireworks/models/minimax-m2p5",
+  "accounts/fireworks/models/minimax-m2p7",
+  "accounts/fireworks/models/deepseek-v4-flash",
+  "accounts/fireworks/models/deepseek-v4-flash-0731",
+  "accounts/fireworks/models/deepseek-v4-flash-vision-exp",
+  "accounts/fireworks/models/deepseek-v4-pro",
+  "accounts/fireworks/models/deepseek-v4-pro-0813",
+  "accounts/fireworks/models/deepseek-v4p1-flash",
+]);
+
+// True when the model accepts `reasoning_history: "preserved"`. Gates the
+// preserve toggle so it never injects a value Fireworks rejects.
+function supportsPreservedReasoningHistory(id: string | undefined): boolean {
+  return !!id && !REASONING_HISTORY_NO_PRESERVE.has(id);
+}
+
 // Runtime state: whether preserved thinking is active. Initialized from the
 // config-file default at session_start (mirrors neuralwatt/makora, which drive
 // preserve state from the config file, not session entries) and updated by the
@@ -1097,6 +1133,7 @@ export {
   setTier,
   updateTierStatus,
   isPreserveEligible,
+  supportsPreservedReasoningHistory,
   setPreserve,
   isValidBiasValue,
   parseLogitBiasMap,
@@ -1142,9 +1179,17 @@ export default function (pi: ExtensionAPI) {
   function notifyPreserveOnSelect(model: any, ctx: any): void {
     if (!model || model.provider !== "fireworks") return;
     if (!isPreserveEligible(model)) return;
-    const msg = preserveOn
-      ? `Preserved thinking ON for ${model.name || model.id} — full reasoning history retained across turns (better multi-turn recall; uses more tokens). Open /fireworks-settings to change.`
-      : `Preserved thinking OFF for ${model.name || model.id} — reasoning stripped each turn (Fireworks default; lighter, weaker multi-turn recall). Open /fireworks-settings to change.`;
+    const name = model.name || model.id;
+    let msg: string;
+    if (preserveOn && !supportsPreservedReasoningHistory(model.id)) {
+      // DeepSeek V4 / MiniMax M2 only support "interleaved"; don't claim
+      // preserved is active for them when it can't be requested.
+      msg = `${name} keeps Fireworks' default reasoning history (interleaved) — preserved thinking is not supported by this model. Open /fireworks-settings to change.`;
+    } else if (preserveOn) {
+      msg = `Preserved thinking ON for ${name} — full reasoning history retained across turns (better multi-turn recall; uses more tokens). Open /fireworks-settings to change.`;
+    } else {
+      msg = `Preserved thinking OFF for ${name} — using the model's default reasoning history (interleaved for most models; lighter, weaker multi-turn recall). Open /fireworks-settings to change.`;
+    }
     if (modelSelectNotifyTimer) clearTimeout(modelSelectNotifyTimer);
     modelSelectNotifyTimer = setTimeout(() => {
       modelSelectNotifyTimer = null;
@@ -1231,12 +1276,14 @@ export default function (pi: ExtensionAPI) {
     // Preserved thinking: inject top-level `reasoning_history: "preserved"`
     // so Fireworks renders prior assistant reasoning (reasoning_content on the
     // OpenAI endpoint, thinking blocks on the Anthropic endpoint) into the
-    // model's context instead of stripping it. The only accepted value is
-    // "preserved"; omitted = stripped (Fireworks default / pi core). Applies to
-    // any Fireworks reasoning model on both transports. pi-ai already replays
-    // the reasoning field/block on prior assistant turns; this flag is what
-    // makes Fireworks honor it. See https://docs.fireworks.ai/guides/reasoning.
-    if (preserveOn && isPreserveEligible(model)) {
+    // model's context instead of stripping it. `reasoning_history` also accepts
+    // "disabled" and "interleaved"; we only ever request the strongest mode and
+    // only on models whose documented support includes it (omitted = the
+    // model/template default, e.g. interleaved for DeepSeek V4). pi-ai already
+    // replays the reasoning field/block on prior assistant turns; this flag is
+    // what makes Fireworks honor it. See
+    // https://docs.fireworks.ai/guides/reasoning#preserved-thinking.
+    if (preserveOn && isPreserveEligible(model) && supportsPreservedReasoningHistory(model.id)) {
       payload.reasoning_history = "preserved";
       modified = true;
     }
@@ -1429,7 +1476,7 @@ export default function (pi: ExtensionAPI) {
           {
             id: "preserveThinking",
             label: "Preserved thinking",
-            description: "Inject reasoning_history:\"preserved\" so Fireworks retains prior assistant reasoning across turns (better multi-turn recall; uses more tokens). Off = Fireworks default (stripped). Applies to every Fireworks reasoning model on both endpoints.",
+            description: "Inject reasoning_history:\"preserved\" so Fireworks retains prior assistant reasoning across turns (better multi-turn recall; uses more tokens). Off = the model's default reasoning history (interleaved for most models). Only sent on models whose documented support includes \"preserved\" (Kimi K2.6/K2.7, GLM 4.7/5.2, …); DeepSeek V4 and MiniMax M2 are interleaved-only.",
             currentValue: preserveOn ? "on" : "off",
             values: ["on", "off"],
           },

package/models.json

@@ -80,6 +80,23 @@
     "contextWindow": 1048576,
     "maxTokens": 0
   },
+  {
+    "id": "accounts/fireworks/models/deepseek-v4p1-flash",
+    "name": "DeepSeek V4.1 Flash",
+    "reasoning": false,
+    "input": [
+      "text",
+      "image"
+    ],
+    "cost": {
+      "input": 0,
+      "output": 0,
+      "cacheRead": 0,
+      "cacheWrite": 0
+    },
+    "contextWindow": 1048576,
+    "maxTokens": 0
+  },
   {
     "id": "accounts/fireworks/models/deepseek-v4-pro",
     "name": "DeepSeek-V4-Pro",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-fireworks-provider",
-  "version": "1.6.2",
+  "version": "1.7.0",
   "description": "Fireworks AI provider extension for pi - Access Kimi, MiniMax, GLM, DeepSeek, and GPT-OSS models through the Fireworks AI API",
   "type": "module",
   "main": "index.ts",

package/patch.json

@@ -15,6 +15,15 @@
     "maxTokens": 163840,
     "compat": {
       "supportsReasoningEffort": true
+    },
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": null,
+      "low": "low",
+      "medium": "medium",
+      "high": "high",
+      "xhigh": null,
+      "max": null
     }
   },
   "accounts/fireworks/models/deepseek-v3p2": {
@@ -33,6 +42,15 @@
     "maxTokens": 160000,
     "compat": {
       "supportsReasoningEffort": true
+    },
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": null,
+      "low": "low",
+      "medium": "medium",
+      "high": "high",
+      "xhigh": null,
+      "max": null
     }
   },
   "accounts/fireworks/models/deepseek-v4-flash": {
@@ -50,7 +68,8 @@
     "contextWindow": 1000000,
     "maxTokens": 384000,
     "compat": {
-      "supportsReasoningEffort": true
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
     },
     "thinkingLevelMap": {
       "off": "none",
@@ -58,7 +77,7 @@
       "low": "low",
       "medium": "medium",
       "high": "high",
-      "xhigh": "xhigh",
+      "xhigh": "max",
       "max": "max"
     }
   },
@@ -77,7 +96,8 @@
     "contextWindow": 1000000,
     "maxTokens": 384000,
     "compat": {
-      "supportsReasoningEffort": true
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
     },
     "thinkingLevelMap": {
       "off": "none",
@@ -85,7 +105,7 @@
       "low": "low",
       "medium": "medium",
       "high": "high",
-      "xhigh": "xhigh",
+      "xhigh": "max",
       "max": "max"
     }
   },
@@ -111,7 +131,7 @@
       "minimal": null,
       "low": null,
       "medium": null,
-      "high": null,
+      "high": "high",
       "xhigh": null,
       "max": null
     }
@@ -138,7 +158,7 @@
       "minimal": null,
       "low": null,
       "medium": null,
-      "high": null,
+      "high": "high",
       "xhigh": null,
       "max": null
     }
@@ -165,7 +185,7 @@
       "minimal": null,
       "low": null,
       "medium": null,
-      "high": null,
+      "high": "high",
       "xhigh": null,
       "max": null
     }
@@ -217,11 +237,11 @@
     "thinkingLevelMap": {
       "off": "none",
       "minimal": null,
-      "low": "high",
-      "medium": "high",
+      "low": null,
+      "medium": null,
       "high": "high",
       "xhigh": null,
-      "max": "max"
+      "max": null
     }
   },
   "accounts/fireworks/models/glm-5p2": {
@@ -235,7 +255,7 @@
       "low": "high",
       "medium": "high",
       "high": "high",
-      "xhigh": null,
+      "xhigh": "max",
       "max": "max"
     },
     "input": [
@@ -378,7 +398,8 @@
     "contextWindow": 262144,
     "maxTokens": 256000,
     "compat": {
-      "supportsReasoningEffort": true
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
     }
   },
   "accounts/fireworks/models/kimi-k2p5": {
@@ -397,7 +418,8 @@
     "contextWindow": 262144,
     "maxTokens": 256000,
     "compat": {
-      "supportsReasoningEffort": true
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
     }
   },
   "accounts/fireworks/models/kimi-k2p6": {
@@ -416,7 +438,8 @@
     "contextWindow": 262000,
     "maxTokens": 262000,
     "compat": {
-      "supportsReasoningEffort": true
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
     }
   },
   "accounts/fireworks/models/kimi-k2p7-code": {
@@ -435,7 +458,8 @@
     "contextWindow": 262000,
     "maxTokens": 262000,
     "compat": {
-      "supportsReasoningEffort": true
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
     }
   },
   "accounts/fireworks/models/minimax-m2p1": {
@@ -798,22 +822,23 @@
       "low": "low",
       "medium": "medium",
       "high": "high",
-      "xhigh": "xhigh",
+      "xhigh": "max",
       "max": "max"
     },
     "input": [
       "text"
     ],
     "cost": {
-      "input": 0.14,
-      "output": 0.28,
-      "cacheRead": 0.028,
+      "input": 0.22,
+      "output": 0.66,
+      "cacheRead": 0.007,
       "cacheWrite": 0
     },
     "contextWindow": 1000000,
     "maxTokens": 384000,
     "compat": {
-      "supportsReasoningEffort": true
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
     }
   },
   "accounts/fireworks/models/inkling": {
@@ -851,12 +876,12 @@
     "contextWindow": 1048576,
     "maxTokens": 131072,
     "thinkingLevelMap": {
-      "off": null,
+      "off": "none",
       "minimal": null,
       "low": "low",
       "medium": "medium",
       "high": "high",
-      "xhigh": null,
+      "xhigh": "max",
       "max": "max"
     },
     "compat": {
@@ -1029,7 +1054,8 @@
     "contextWindow": 1048576,
     "maxTokens": 384000,
     "compat": {
-      "supportsReasoningEffort": true
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
     },
     "thinkingLevelMap": {
       "off": "none",
@@ -1037,8 +1063,14 @@
       "low": "low",
       "medium": "medium",
       "high": "high",
-      "xhigh": "xhigh",
+      "xhigh": "max",
       "max": "max"
+    },
+    "cost": {
+      "input": 1.32,
+      "output": 3.96,
+      "cacheRead": 0.044,
+      "cacheWrite": 0
     }
   },
   "accounts/fireworks/models/muse-glimmer-30b": {
@@ -1052,6 +1084,12 @@
     "maxTokens": 131072,
     "compat": {
       "supportsReasoningEffort": true
+    },
+    "cost": {
+      "input": 0.35,
+      "output": 1.5,
+      "cacheRead": 0.04,
+      "cacheWrite": 0
     }
   },
   "accounts/fireworks/models/nemotron-lightning-3p5-30b-a3b": {
@@ -1064,6 +1102,12 @@
     "maxTokens": 262144,
     "compat": {
       "supportsReasoningEffort": true
+    },
+    "cost": {
+      "input": 0.05,
+      "output": 0.2,
+      "cacheRead": 0.01,
+      "cacheWrite": 0
     }
   },
   "accounts/fireworks/models/qwen3p8-2p4t-a95b": {
@@ -1080,8 +1124,93 @@
       "low": "low",
       "medium": "medium",
       "high": "high",
-      "xhigh": null,
-      "max": null
+      "xhigh": "xhigh",
+      "max": "max"
+    },
+    "compat": {
+      "supportsReasoningEffort": true
+    }
+  },
+  "accounts/fireworks/models/deepseek-v4p1-flash": {
+    "name": "DeepSeek V4.1 Flash",
+    "reasoning": true,
+    "input": [
+      "text",
+      "image"
+    ],
+    "cost": {
+      "input": 0.22,
+      "output": 0.66,
+      "cacheRead": 0.007,
+      "cacheWrite": 0
+    },
+    "contextWindow": 1048576,
+    "maxTokens": 384000,
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": null,
+      "low": "low",
+      "medium": "medium",
+      "high": "high",
+      "xhigh": "max",
+      "max": "max"
+    },
+    "compat": {
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
+    }
+  },
+  "accounts/fireworks/models/deepseek-v4-flash-vision-exp": {
+    "name": "DeepSeek V4 Flash Vision Exp",
+    "reasoning": true,
+    "input": [
+      "text",
+      "image"
+    ],
+    "cost": {
+      "input": 0.22,
+      "output": 0.66,
+      "cacheRead": 0.007,
+      "cacheWrite": 0
+    },
+    "contextWindow": 1048576,
+    "maxTokens": 384000,
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": null,
+      "low": "low",
+      "medium": "medium",
+      "high": "high",
+      "xhigh": "max",
+      "max": "max"
+    },
+    "compat": {
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
+    }
+  },
+  "accounts/fireworks/models/qwen3p8-max": {
+    "name": "Qwen3.8 Max",
+    "reasoning": true,
+    "input": [
+      "text"
+    ],
+    "cost": {
+      "input": 2,
+      "output": 6,
+      "cacheRead": 0.25,
+      "cacheWrite": 0
+    },
+    "contextWindow": 262144,
+    "maxTokens": 131072,
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": null,
+      "low": "low",
+      "medium": "medium",
+      "high": "high",
+      "xhigh": "xhigh",
+      "max": "max"
     },
     "compat": {
       "supportsReasoningEffort": true