loom-agent 1.5.0 → 1.6.0

package/CHANGELOG.md

@@ -4,7 +4,9 @@ All notable changes to **Loom Code** are documented here.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
-## [Unreleased] — completion harness, specialist agents, headless browser, TUI refresh
+## [Unreleased]
+
+## [1.6.0] — 2026-10-17
 
 ### Added
 - **Transcript undo / redo (`/undo`, `/redo`, `/undo-files`).** The
@@ -38,6 +40,50 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   `<leader>x` keybind, which previously resolved to a slash command with no
   handler.
 
+### Fixed
+- **Model catalogs verified against live provider APIs.** Every model ID the
+  picker ships was checked against the authoritative source (OpenAI,
+  NVIDIA, OpenRouter and Groq `/v1/models`; Anthropic's official pricing +
+  deprecation pages; the models.dev cache for registry providers), and every
+  dead ID was removed or repointed:
+  - **NVIDIA** (the reported failure): the default active model
+    `nvidia/llama-3.3-nemotron-super-49b-v1.5` did not exist and 404'd on
+    every request — repointed to `nvidia/nemotron-3-super-120b-a12b`
+    (54/54 catalog models live).
+  - **OpenAI:** dropped phantom `gpt-5.6`, `gpt-5.3-codex-spark`, `o3-pro`;
+    added live `gpt-5-codex`, `gpt-5.1-codex`, `gpt-5.1-codex-max`,
+    `gpt-5.1-codex-mini`, `gpt-5.2-codex`, `gpt-5.2-pro` (34/34 live).
+  - **Anthropic:** dropped models Anthropic retired in 2026
+    (`claude-sonnet-4-20250514`, `claude-opus-4-20250514`,
+    `claude-3-5-haiku-20241022`); added live `claude-fable-5`,
+    `claude-opus-4-7`, `claude-opus-4-6` (14 models, all prices matched to
+    the official pricing table; Mythos 5/5.1 omitted — verification-program
+    accounts only).
+  - **Defaults:** `settings.js` DEFAULTS and the shipped model map no longer
+    name dead IDs (`gpt-5-fast`, `meta/llama-3.1-8b-instruct`,
+    `anthropic/claude-sonnet-4`, `moonshotai/kimi-k3-free` → `gpt-5`,
+    `nemotron-3-super-120b-a12b`, `claude-sonnet-5-5`,
+    `z-ai/glm-5.3-free`); OpenRouter 42/42 and TokenRouter (models.dev's
+    single `z-ai/glm-5.3-free`) verified live.
+  - Usage/context tests now pin a deterministic model/window so future
+    default-model changes can't shift their expectations.
+- **Speed telemetry now ticks during the thinking phase.** Reasoning deltas
+  count as first-token + generation tokens, so the sidebar latency gauge
+  starts when the model starts *thinking* instead of reading "waiting…"
+  through the entire thinking phase (measured on nemotron-3-super: ~7s to
+  the first thinking token, then +16s of thought before any visible text —
+  the response looked hung while the model was working). `last.latencyMs`
+  and `last.tokensPerSec` now reflect true time-to-first-model-output and
+  full generation throughput, including reasoning tokens. Regression test
+  in `session.test.js`.
+- **NVIDIA catalog note: `/v1/models` is not account availability.** The
+  endpoint lists NVIDIA's global catalog, but chat POSTs can return
+  `Function '…': Not found for account '…'` when the key's account hasn't
+  enabled that model (probed 2026-10-17: 9/54 curated IDs answered chat on
+  one key — 37 account-gated 404s, ~6 hung to timeout, 2 returned 503).
+  Availability is per-account, so the curated list stays; the caveat is
+  now documented in `nvidia.js`.
+
 ### Changed
 - **Orchestrator now delegates by default.** The build agent's standing
   instructions were rewritten to make fan-out the first move on any broad

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "loom-agent",
-  "version": "1.5.0",
+  "version": "1.6.0",
   "repository": {
     "type": "git",
     "url": "git+https://github.com/toshalkumbhar8979-design/loomcode.git"

package/src/config/settings.js

@@ -20,12 +20,17 @@ function configFile() {
 const DEFAULTS = {
   provider: 'anthropic',
   model: {
-    anthropic: 'claude-sonnet-4-20250514',
-    openai: 'gpt-5-fast',
-    nvidia: 'meta/llama-3.1-8b-instruct',
+    // claude-sonnet-4-20250514 was retired 2026-06-15 (404); Anthropic's
+    // recommended replacement is claude-sonnet-5-5.
+    anthropic: 'claude-sonnet-5-5',
+    // Defaults must be IDs that exist on the LIVE provider APIs (verified
+    // 2026-10-17): gpt-5-fast and meta/llama-3.1-8b-instruct returned 404,
+    // anthropic/claude-sonnet-4 and moonshotai/kimi-k3-free are gone too.
+    openai: 'gpt-5',
+    nvidia: 'nvidia/nemotron-3-super-120b-a12b',
     google: 'gemini-2.5-flash',
-    openrouter: 'anthropic/claude-sonnet-4',
-    tokenrouter: 'moonshotai/kimi-k3-free',
+    openrouter: 'anthropic/claude-sonnet-5.5',
+    tokenrouter: 'z-ai/glm-5.3-free',
     local: 'llama3.2',
   },
   maxTokens: 16384,

package/src/core/session.js

@@ -696,7 +696,16 @@ ${this.mode === 'build' ? `- You have all tools available: read, write, edit, ba
           if (!speed._firstTokenAt) speed._firstTokenAt = now;
           if (callbacks.onDelta) callbacks.onDelta(txt);
         }, onReasoning: (txt) => {
+          const now = Date.now();
           reasoning += txt;
+          // Thinking tokens are still model output: count them toward the
+          // live gauge so first-token latency ticks from the first THINKING
+          // token. Otherwise the sidebar reads "waiting…" for the entire
+          // thinking phase (measured: ~7s to first thinking delta, then +16s
+          // of thought before any visible text on nemotron-3-super) and the
+          // response looks hung while the model is actually working.
+          speed._liveTokens += txt.length / 4;
+          if (!speed._firstTokenAt) speed._firstTokenAt = now;
           if (callbacks.onReasoning) callbacks.onReasoning(txt);
         } }
       : callbacks;
@@ -838,7 +847,7 @@ ${this.mode === 'build' ? `- You have all tools available: read, write, edit, ba
       // Finalize the speed snapshot for this model call (or tool step).
       if (resp && !resp.interrupted) {
         const dur = Date.now() - speed._turnStart;
-        const tokens = (streamed.length || String(resp.content || '').length) / 4;
+        const tokens = ((streamed + reasoning).length || String(resp.content || '').length) / 4;
         speed.lastDurationMs = dur;
         speed.lastTokens = Math.round(tokens);
         speed.lastTokensPerSec = dur > 0 ? Math.round((tokens / dur) * 1000) : 0;

package/src/providers/anthropic.js

@@ -328,6 +328,9 @@ const models = [
   { id: 'claude-opus-4-8', name: 'Claude Opus 4.8', provider: 'anthropic', context: 1000000, priceIn: 5, priceOut: 25, tags: ['frontier', 'reasoning'] },
   { id: 'claude-opus-4-5', name: 'Claude Opus 4.5', provider: 'anthropic', context: 200000, priceIn: 5, priceOut: 25, tags: ['frontier', 'reasoning'] },
   { id: 'claude-fable-5-1', name: 'Claude Fable 5.1', provider: 'anthropic', context: 1000000, priceIn: 10, priceOut: 50, tags: ['frontier', 'reasoning'] },
+  { id: 'claude-fable-5', name: 'Claude Fable 5', provider: 'anthropic', context: 1000000, priceIn: 10, priceOut: 50, tags: ['frontier', 'reasoning'] },
+  { id: 'claude-opus-4-7', name: 'Claude Opus 4.7', provider: 'anthropic', context: 1000000, priceIn: 5, priceOut: 25, tags: ['frontier', 'reasoning'] },
+  { id: 'claude-opus-4-6', name: 'Claude Opus 4.6', provider: 'anthropic', context: 1000000, priceIn: 5, priceOut: 25, tags: ['frontier', 'reasoning'] },
   // ── Sonnet ──
   { id: 'claude-sonnet-5-5', name: 'Claude Sonnet 5.5', provider: 'anthropic', context: 1000000, priceIn: 2, priceOut: 10, tags: ['reasoning'] },
   { id: 'claude-sonnet-5', name: 'Claude Sonnet 5', provider: 'anthropic', context: 1000000, priceIn: 2, priceOut: 10, tags: ['reasoning'] },
@@ -337,9 +340,11 @@ const models = [
   { id: 'claude-haiku-5-5', name: 'Claude Haiku 5.5', provider: 'anthropic', context: 1000000, priceIn: 0.1, priceOut: 0.5, tags: ['fast', 'cheap', 'reasoning'] },
   { id: 'claude-haiku-4-5', name: 'Claude Haiku 4.5', provider: 'anthropic', context: 200000, priceIn: 1, priceOut: 5, tags: ['fast', 'reasoning'] },
   // ── Legacy snapshots ──
-  { id: 'claude-sonnet-4-20250514', name: 'Claude Sonnet 4', provider: 'anthropic', context: 200000, priceIn: 3, priceOut: 15, tags: ['reasoning', 'legacy'] },
-  { id: 'claude-opus-4-20250514', name: 'Claude Opus 4', provider: 'anthropic', context: 200000, priceIn: 15, priceOut: 75, tags: ['reasoning', 'legacy'] },
-  { id: 'claude-3-5-haiku-20241022', name: 'Claude 3.5 Haiku', provider: 'anthropic', context: 200000, priceIn: 0.8, priceOut: 4, tags: ['legacy'] },
+  // Retired models verified against Anthropic's model-deprecations page
+  // (2026-10-17): claude-sonnet-4-20250514 and claude-opus-4-20250514 were
+  // retired 2026-06-15, claude-3-5-haiku-20241022 retired 2026-02-19 — they
+  // 404 and are no longer listed. claude-mythos-5/5.1 exist but are limited
+  // to verified organizations, so they are not offered in the picker.
 ];
 
 module.exports = { chat, stream, models, buildBody, normalizeBlocks };

package/src/providers/nvidia.js

@@ -13,6 +13,15 @@ function getKey() {
 // deepseek-ai/deepseek-v4-flash-0731, z-ai/glm-5.2, openai/gpt-oss-120b —
 // return 404/410 and were removed. Re-verify against /v1/models before
 // adding any ID.
+//
+// NOTE (probed 2026-10-17): GET /v1/models is the GLOBAL catalog — a chat
+// POST can still fail with `Function '<id>': Not found for account '<key>'`,
+// i.e. the model exists but the key's account hasn't enabled it. On one
+// sample key only 9/54 of these IDs answered a chat call (37 account-gated
+// 404s, ~6 hang to timeout, 2 return 503 capacity errors), so the picker
+// may offer models that 404 for a given account. Availability is
+// per-account, so the full curated list stays; NVIDIA's own errors carry
+// the account hint.
 const provider = {
   ...createOpenAICompatProvider({
     getKey,

package/src/providers/openai.js

@@ -18,7 +18,6 @@ const provider = {
   { id: 'gpt-6-luna', name: 'GPT-6 Luna', provider: 'openai', tags: ['fast', 'cheap', 'reasoning'], context: 1050000, priceIn: 0.1, priceOut: 0.5 },
 
   // ── GPT-5.6 / 5.5 / 5.4 ──
-  { id: 'gpt-5.6', name: 'GPT-5.6', provider: 'openai', tags: ['frontier', 'reasoning'], context: 1050000, priceIn: 4, priceOut: 20 },
   { id: 'gpt-5.6-sol', name: 'GPT-5.6 Sol', provider: 'openai', tags: ['reasoning'], context: 1050000, priceIn: 4, priceOut: 20 },
   { id: 'gpt-5.6-terra', name: 'GPT-5.6 Terra', provider: 'openai', tags: ['reasoning'], context: 1050000, priceIn: 2, priceOut: 12 },
   { id: 'gpt-5.6-luna', name: 'GPT-5.6 Luna', provider: 'openai', tags: ['fast', 'cheap', 'reasoning'], context: 1050000, priceIn: 0.2, priceOut: 1.2 },
@@ -31,10 +30,15 @@ const provider = {
 
   // ── GPT-5 Codex (coding) ──
   { id: 'gpt-5.3-codex', name: 'GPT-5.3 Codex', provider: 'openai', tags: ['coding', 'reasoning'], context: 400000, priceIn: 1.75, priceOut: 14 },
-  { id: 'gpt-5.3-codex-spark', name: 'GPT-5.3 Codex Spark', provider: 'openai', tags: ['coding', 'fast', 'reasoning'], context: 128000, priceIn: 1.75, priceOut: 14 },
+  { id: 'gpt-5.2-codex', name: 'GPT-5.2 Codex', provider: 'openai', tags: ['coding', 'reasoning'], context: 400000, priceIn: 1.75, priceOut: 14 },
+  { id: 'gpt-5.1-codex-max', name: 'GPT-5.1 Codex Max', provider: 'openai', tags: ['coding', 'reasoning'], context: 400000, priceIn: 1.75, priceOut: 14 },
+  { id: 'gpt-5.1-codex', name: 'GPT-5.1 Codex', provider: 'openai', tags: ['coding', 'reasoning'], context: 400000, priceIn: 1.25, priceOut: 10 },
+  { id: 'gpt-5.1-codex-mini', name: 'GPT-5.1 Codex Mini', provider: 'openai', tags: ['coding', 'fast', 'reasoning'], context: 400000, priceIn: 0.25, priceOut: 2 },
+  { id: 'gpt-5-codex', name: 'GPT-5 Codex', provider: 'openai', tags: ['coding', 'reasoning'], context: 400000, priceIn: 1.25, priceOut: 10 },
 
   // ── GPT-5 Series ──
   { id: 'gpt-5.2', name: 'GPT-5.2', provider: 'openai', tags: ['reasoning'], context: 400000, priceIn: 1.75, priceOut: 14 },
+  { id: 'gpt-5.2-pro', name: 'GPT-5.2 Pro', provider: 'openai', tags: ['reasoning', 'pro'], context: 400000, priceIn: 21, priceOut: 168 },
   { id: 'gpt-5.1', name: 'GPT-5.1', provider: 'openai', tags: ['reasoning'], context: 400000, priceIn: 1.25, priceOut: 10 },
   { id: 'gpt-5', name: 'GPT-5', provider: 'openai', tags: ['frontier', 'reasoning'], context: 400000, priceIn: 1.25, priceOut: 10 },
   { id: 'gpt-5-mini', name: 'GPT-5 Mini', provider: 'openai', tags: ['fast', 'reasoning'], context: 400000, priceIn: 0.25, priceOut: 2 },
@@ -43,7 +47,6 @@ const provider = {
 
   // ── O-Series (Reasoning, legacy) ──
   { id: 'o3', name: 'o3', provider: 'openai', tags: ['reasoning', 'legacy'], context: 200000, priceIn: 2, priceOut: 8 },
-  { id: 'o3-pro', name: 'o3 Pro', provider: 'openai', tags: ['reasoning', 'pro', 'legacy'], context: 200000, priceIn: 20, priceOut: 80 },
   { id: 'o3-mini', name: 'o3 Mini', provider: 'openai', tags: ['reasoning', 'fast', 'legacy'], context: 200000, priceIn: 1.1, priceOut: 4.4 },
   { id: 'o4-mini', name: 'o4 Mini', provider: 'openai', tags: ['reasoning', 'fast', 'legacy'], context: 200000, priceIn: 1.1, priceOut: 4.4 },
 

package/src/providers/tokenrouter.js

@@ -12,7 +12,10 @@ const provider = {
     envKeyHint: 'TOKENROUTER',
   }),
   models: [
-  { id: 'moonshotai/kimi-k3-free', name: 'Kimi K3 Free', provider: 'tokenrouter', tags: ['free'], context: 128000, priceIn: 0, priceOut: 0 },
+  // models.dev (2026-10-17): TokenRouter's only listed model is the free
+  // GLM-5.3 endpoint; the earlier moonshotai/kimi-k3-free ID is gone and
+  // returns an error. Verified against the models.dev registry cache.
+  { id: 'z-ai/glm-5.3-free', name: 'GLM 5.3 Free', provider: 'tokenrouter', tags: ['free'], context: 1000000, priceIn: 0, priceOut: 0 },
   ],
 };
 

package/src/tui/components/SubagentPanel.tsx

@@ -110,7 +110,7 @@ export function SubagentPanel() {
       const curModel = (cfg.agents[cur.agentId] && cfg.agents[cur.agentId].model) || "";
       openModal({
         type: "input", title: "Default model for " + cur.agent,
-        placeholder: 'e.g. "anthropic/claude-sonnet-4-20250514" — empty clears',
+        placeholder: 'e.g. "anthropic/claude-sonnet-5-5" — empty clears',
         value: curModel,
         onPick(val: string) {
           closeModal();