feat: add Claude Fable 5 and sync Anthropic models from the first-party Models API#733
Open
tombeckenham wants to merge 4 commits into
Open
feat: add Claude Fable 5 and sync Anthropic models from the first-party Models API#733tombeckenham wants to merge 4 commits into
tombeckenham wants to merge 4 commits into
Conversation
…adapters Adds claude-fable-5 to scripts/openrouter.models.json and regenerates the model lists via the codegen pipeline (regenerate:models + sync-provider-models), registering the model in both adapters' model metadata, provider option types, and input modality types. https://claude.ai/code/session_01GFGB3ZPVG4tMpYnUYUVpPK
…f OpenRouter Adds scripts/fetch-anthropic-models.ts (snapshots GET /v1/models to scripts/anthropic.models.json) and scripts/sync-anthropic-models.ts (inserts missing models into packages/ai-anthropic/src/model-meta.ts). The Models API is authoritative for model IDs, context windows, max output tokens, and per-model capabilities (adaptive vs extended thinking, image input) — unlike OpenRouter slugs, which use dotted IDs that are not valid Anthropic API model IDs. Pricing is not exposed by the Models API, so it lives in a hand-maintained PRICING table and the sync fails loudly for new models without an entry. The 'anthropic/' provider is removed from sync-provider-models.ts so Anthropic models are no longer added from OpenRouter data. New package.json scripts: generate:anthropic-models[:fetch], regenerate:anthropic-models. Also corrects claude-fable-5 thinking flags (extended_thinking: false, adaptive_thinking: true) to match what the capability-aware sync generates. https://claude.ai/code/session_01GFGB3ZPVG4tMpYnUYUVpPK
Contributor
📝 WalkthroughWalkthroughThis PR adds Claude Fable 5 model support across Anthropic and OpenRouter adapters, corrects Anthropic model metadata from live APIs (pricing, context windows), fixes model ID formatting from dotted to hyphenated notation, and establishes a new sync pipeline that sources Anthropic models from the first-party Models API instead of OpenRouter. ChangesClaude Fable 5 Model Support & Anthropic First-Party Sync Pipeline
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Possibly related PRs
Suggested reviewers
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
View your CI Pipeline Execution ↗ for commit 2e15251
☁️ Nx Cloud last updated this comment at |
@tanstack/ai
@tanstack/ai-anthropic
@tanstack/ai-client
@tanstack/ai-code-mode
@tanstack/ai-code-mode-skills
@tanstack/ai-devtools-core
@tanstack/ai-elevenlabs
@tanstack/ai-event-client
@tanstack/ai-fal
@tanstack/ai-gemini
@tanstack/ai-grok
@tanstack/ai-groq
@tanstack/ai-isolate-cloudflare
@tanstack/ai-isolate-node
@tanstack/ai-isolate-quickjs
@tanstack/ai-mcp
@tanstack/ai-ollama
@tanstack/ai-openai
@tanstack/ai-openrouter
@tanstack/ai-preact
@tanstack/ai-react
@tanstack/ai-react-ui
@tanstack/ai-solid
@tanstack/ai-solid-ui
@tanstack/ai-svelte
@tanstack/ai-utils
@tanstack/ai-vue
@tanstack/ai-vue-ui
@tanstack/openai-base
@tanstack/preact-ai-devtools
@tanstack/react-ai-devtools
@tanstack/solid-ai-devtools
commit: |
…e Models API - Fix claude-opus-4-5 pricing ($5/$25, was wrongly $15/$75) in the sync script's PRICING table and model-meta.ts; add documented cache-read prices to all legacy model entries - Rename invalid OpenRouter-style ids claude-opus-4.8/-fast to the real Anthropic API ids claude-opus-4-8/-fast (dotted ids 404), including three docs pages that copied them - Correct stale context windows / output limits per the Models API (Opus 4.6, Sonnet 4.5, Sonnet 4: 1M ctx; Sonnet 4.6: 128K out; Opus 4.5: 64K out; Opus 4.1: 32K out) - Register claude-fable-5, claude-opus-4-7-fast, claude-opus-4-8 and claude-opus-4-8-fast in AnthropicChatModelToolCapabilitiesByName, teach the sync script to maintain that map, and add a drift-guard type test - Refresh scripts/anthropic.models.json from a genuine live fetch (11 models, was a hand-trimmed 6) - Bump @anthropic-ai/sdk to ^0.104.0 (Model union includes claude-fable-5 and claude-opus-4-8) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…orkflow pnpm generate:models now chains the Anthropic fetch + sync after the OpenRouter sync, so the nightly sync-models workflow keeps @tanstack/ai-anthropic current too. The workflow passes ANTHROPIC_API_KEY (new repo secret required) to the sync step and commits the refreshed scripts/anthropic.models.json snapshot alongside package changes. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
Review the following changes in direct dependencies. Learn more about Socket for GitHub.
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (2)
.github/workflows/sync-models.yml (1)
40-44:⚠️ Potential issue | 🟠 Major | ⚡ Quick winChange detection is too narrow for the files this job now writes.
At Line 40, the gate only checks
packages/, but this workflow also updatesscripts/anthropic.models.json,scripts/openrouter.models.json,scripts/.sync-models-last-run, and.changeset/. If only those change, commit/PR creation is skipped.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In @.github/workflows/sync-models.yml around lines 40 - 44, The git-change gate currently checks only the command "git diff --quiet -- packages/" so the job will miss changes made to scripts/anthropic.models.json, scripts/openrouter.models.json, scripts/.sync-models-last-run, and .changeset/; update the condition that writes "changed=true/false" to include those paths (e.g., extend the git diff check to include scripts/, .changeset/, and the specific model files) or switch to a broader check (e.g., any changes outside the excluded set) so changes to scripts/ and .changeset/ cause the job to run and create the commit/PR; locate the existing git diff check string "git diff --quiet -- packages/" in the workflow and modify it accordingly.packages/ai-anthropic/src/model-meta.ts (1)
794-803:⚠️ Potential issue | 🟡 MinorAdd
CLAUDE_FABLE_5toANTHROPIC_COMBINED_TOOLS_AND_SCHEMA_MODELS
claude-fable-5supports structured outputs viaoutput_config.format, and it can be used together with tool calls; without being in this Set, the adapter won’t opt into the native combined tools+schema streaming path and will fall back to the legacystructuredOutputfinalization flow instead.Proposed diff
export const ANTHROPIC_COMBINED_TOOLS_AND_SCHEMA_MODELS = new Set<string>([ CLAUDE_OPUS_4_5.id, CLAUDE_OPUS_4_6.id, CLAUDE_OPUS_4_6_FAST.id, CLAUDE_OPUS_4_7.id, CLAUDE_OPUS_4_7_FAST.id, CLAUDE_SONNET_4_5.id, CLAUDE_SONNET_4_6.id, CLAUDE_HAIKU_4_5.id, + CLAUDE_FABLE_5.id, ])🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@packages/ai-anthropic/src/model-meta.ts` around lines 794 - 803, Add CLAUDE_FABLE_5.id to the ANTHROPIC_COMBINED_TOOLS_AND_SCHEMA_MODELS Set so the adapter opts into the native combined tools+schema streaming path; locate the Set declaration named ANTHROPIC_COMBINED_TOOLS_AND_SCHEMA_MODELS and append CLAUDE_FABLE_5.id alongside the other model ids (e.g., CLAUDE_OPUS_4_5.id, CLAUDE_SONNET_4_6.id) to ensure claude-fable-5 uses output_config.format and combined tools+schema streaming.
🧹 Nitpick comments (2)
scripts/sync-anthropic-models.ts (1)
313-347: ⚡ Quick winSort
newModelsbefore generation to keep diffs deterministic.At Line 313+, emitted constants/map entries currently follow map insertion order from the snapshot. Sorting by alias before Line 345 reduces diff churn across runs.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@scripts/sync-anthropic-models.ts` around lines 313 - 347, The newModels array is built from byAlias and preserves snapshot insertion order causing non-deterministic diffs; sort newModels by alias (e.g., using localeCompare on the alias field) after you finish populating it and before you log/print and before you generate constants with generateModelConstant and toConstName so generated map entries and constants are emitted in a stable, deterministic order; ensure the sort happens prior to the check/newModels.length logging and the constants = newModels.map(...) call.packages/ai-anthropic/tests/model-meta.test.ts (1)
2-2: ⚡ Quick winUse
import typefor type-only imports.The
ANTHROPIC_MODELSimport on line 2 is only used in type positions (line 253:typeof ANTHROPIC_MODELS). Per the project's TypeScript conventions and ESLint rules, this should useimport typeinstead of a value import.♻️ Suggested fix
-import { ANTHROPIC_MODELS } from '../src/model-meta' +import type { ANTHROPIC_MODELS } from '../src/model-meta'🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@packages/ai-anthropic/tests/model-meta.test.ts` at line 2, The import of ANTHROPIC_MODELS in packages/ai-anthropic/tests/model-meta.test.ts is only used in type positions (e.g., typeof ANTHROPIC_MODELS) so change the value import to a type-only import: replace the current import of ANTHROPIC_MODELS with an import type statement (import type { ANTHROPIC_MODELS } from '../src/model-meta') so TypeScript/ESLint treat it as a type-only import and avoid emitting runtime imports.Source: Linters/SAST tools
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@packages/ai-anthropic/src/model-meta.ts`:
- Around line 722-760: CLAUDE_FABLE_5 is typed with AnthropicThinkingOptions but
the model only supports adaptive/effort controls (extended_thinking: false);
remove AnthropicThinkingOptions from the generic type list for CLAUDE_FABLE_5
and replace it with the specific adaptive/effort options type used by your
codebase (e.g., AnthropicAdaptiveThinkingOptions or AnthropicEffortOptions) so
the provider-options reflect effort/`effort`-style controls instead of the
`thinking: { type: 'enabled', budget_tokens }` shape; update the satisfies
ModelMeta<...> clause for CLAUDE_FABLE_5 accordingly and ensure imports/types
are adjusted to reference the adaptive/effort options symbol.
In `@scripts/fetch-anthropic-models.ts`:
- Around line 39-44: The fetch call in scripts/fetch-anthropic-models.ts (the
const response = await fetch(url, { ... }) line) needs an abort timeout to avoid
hanging; create an AbortController (or use AbortSignal.timeout) and pass its
signal in the fetch options, set a sensible timeout (e.g., 10s) and ensure you
handle the abort by catching the thrown error and logging/propagating it back
from the function that calls fetch; keep the existing headers ('x-api-key' and
'anthropic-version') and ensure any manual timer is cleared on success when
using AbortController.
---
Outside diff comments:
In @.github/workflows/sync-models.yml:
- Around line 40-44: The git-change gate currently checks only the command "git
diff --quiet -- packages/" so the job will miss changes made to
scripts/anthropic.models.json, scripts/openrouter.models.json,
scripts/.sync-models-last-run, and .changeset/; update the condition that writes
"changed=true/false" to include those paths (e.g., extend the git diff check to
include scripts/, .changeset/, and the specific model files) or switch to a
broader check (e.g., any changes outside the excluded set) so changes to
scripts/ and .changeset/ cause the job to run and create the commit/PR; locate
the existing git diff check string "git diff --quiet -- packages/" in the
workflow and modify it accordingly.
In `@packages/ai-anthropic/src/model-meta.ts`:
- Around line 794-803: Add CLAUDE_FABLE_5.id to the
ANTHROPIC_COMBINED_TOOLS_AND_SCHEMA_MODELS Set so the adapter opts into the
native combined tools+schema streaming path; locate the Set declaration named
ANTHROPIC_COMBINED_TOOLS_AND_SCHEMA_MODELS and append CLAUDE_FABLE_5.id
alongside the other model ids (e.g., CLAUDE_OPUS_4_5.id, CLAUDE_SONNET_4_6.id)
to ensure claude-fable-5 uses output_config.format and combined tools+schema
streaming.
---
Nitpick comments:
In `@packages/ai-anthropic/tests/model-meta.test.ts`:
- Line 2: The import of ANTHROPIC_MODELS in
packages/ai-anthropic/tests/model-meta.test.ts is only used in type positions
(e.g., typeof ANTHROPIC_MODELS) so change the value import to a type-only
import: replace the current import of ANTHROPIC_MODELS with an import type
statement (import type { ANTHROPIC_MODELS } from '../src/model-meta') so
TypeScript/ESLint treat it as a type-only import and avoid emitting runtime
imports.
In `@scripts/sync-anthropic-models.ts`:
- Around line 313-347: The newModels array is built from byAlias and preserves
snapshot insertion order causing non-deterministic diffs; sort newModels by
alias (e.g., using localeCompare on the alias field) after you finish populating
it and before you log/print and before you generate constants with
generateModelConstant and toConstName so generated map entries and constants are
emitted in a stable, deterministic order; ensure the sort happens prior to the
check/newModels.length logging and the constants = newModels.map(...) call.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: b141ad52-b57a-4284-9625-9895e82ce829
⛔ Files ignored due to path filters (1)
pnpm-lock.yamlis excluded by!**/pnpm-lock.yaml
📒 Files selected for processing (16)
.changeset/fable-model-support.md.github/workflows/sync-models.ymldocs/adapters/anthropic.mddocs/advanced/multimodal-content.mddocs/tools/provider-tools.mdpackage.jsonpackages/ai-anthropic/package.jsonpackages/ai-anthropic/src/model-meta.tspackages/ai-anthropic/tests/model-meta.test.tspackages/ai-openrouter/src/model-meta.tsscripts/.sync-models-last-runscripts/anthropic.models.jsonscripts/fetch-anthropic-models.tsscripts/openrouter.models.jsonscripts/sync-anthropic-models.tsscripts/sync-provider-models.ts
Comment on lines
+722
to
+760
| const CLAUDE_FABLE_5 = { | ||
| name: 'claude-fable-5', | ||
| id: 'claude-fable-5', | ||
| context_window: 1_000_000, | ||
| max_output_tokens: 128_000, | ||
| supports: { | ||
| input: ['text', 'image', 'document'], | ||
| extended_thinking: false, | ||
| adaptive_thinking: true, | ||
| priority_tier: true, | ||
| tools: [ | ||
| 'web_search', | ||
| 'web_fetch', | ||
| 'code_execution', | ||
| 'computer_use', | ||
| 'bash', | ||
| 'text_editor', | ||
| 'memory', | ||
| ], | ||
| }, | ||
| pricing: { | ||
| input: { | ||
| normal: 10, | ||
| cached: 1, | ||
| }, | ||
| output: { | ||
| normal: 50, | ||
| }, | ||
| }, | ||
| } as const satisfies ModelMeta< | ||
| AnthropicContainerOptions & | ||
| AnthropicContextManagementOptions & | ||
| AnthropicMCPOptions & | ||
| AnthropicServiceTierOptions & | ||
| AnthropicStopSequencesOptions & | ||
| AnthropicThinkingOptions & | ||
| AnthropicToolChoiceOptions & | ||
| AnthropicSamplingOptions | ||
| > |
Contributor
There was a problem hiding this comment.
🧩 Analysis chain
🌐 Web query:
Does Anthropic's Claude Fable 5 model support extended thinking or thinking budget configuration via the thinking parameter?
💡 Result:
Anthropic's Claude Fable 5 model does not support manual extended thinking configuration via the traditional thinking parameter and budget_tokens [1][2][3]. Instead, Claude Fable 5 uses "adaptive thinking," which is permanently enabled and cannot be disabled [1][3]. Because thinking is always active, attempting to use the manual thinking: {type: "enabled", budget_tokens: N} configuration will result in a 400 error [2][3]. Similarly, attempting to disable it with thinking: {type: "disabled"} is not supported and will also trigger an error [2]. In place of a manual token budget, you manage the trade-off between intelligence, latency, and cost on Claude Fable 5 using the effort parameter [4][3]. The effort parameter allows you to guide the model's adaptive thinking process, with options such as high (recommended default), xhigh (for maximum capability), and medium or low (for routine tasks) [4].
Citations:
- 1: https://platform.claude.com/docs/en/about-claude/models/overview
- 2: https://platform.claude.com/docs/en/build-with-claude/extended-thinking
- 3: https://platform.claude.com/docs/en/build-with-claude/adaptive-thinking
- 4: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5
🏁 Script executed:
#!/bin/bash
set -euo pipefail
FILE="packages/ai-anthropic/src/model-meta.ts"
# Show the Fable 5 block with nearby type intersections
sed -n '700,790p' "$FILE" | nl -ba
# Show the Haiku 3.5 and Haiku 3 blocks that were referenced
sed -n '860,940p' "$FILE" | nl -ba
# Search for AnthropicThinkingOptions usage within this file
rg -n "AnthropicThinkingOptions" "$FILE"Repository: TanStack/ai
Length of output: 98
🏁 Script executed:
#!/bin/bash
set -euo pipefail
FILE="packages/ai-anthropic/src/model-meta.ts"
echo "== Fable 5 block (approx 700-790) =="
sed -n '700,790p' "$FILE" | awk '{printf "%5d\t%s\n", NR, $0}'
echo
echo "== Haiku blocks (approx 860-940) =="
sed -n '860,940p' "$FILE" | awk '{printf "%5d\t%s\n", NR, $0}'
echo
echo "== All AnthropicThinkingOptions occurrences =="
rg -n "AnthropicThinkingOptions" "$FILE" || trueRepository: TanStack/ai
Length of output: 7131
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# Locate definition(s) of AnthropicThinkingOptions and what parameters it covers
rg -n "type\s+AnthropicThinkingOptions|interface\s+AnthropicThinkingOptions|export\s+type\s+AnthropicThinkingOptions|export\s+interface\s+AnthropicThinkingOptions" packages/ai-anthropic/src/model-meta.ts packages/ai-anthropic/src -S || true
echo "---- Search for budget_tokens / effort in the repo (ai-anthropic) ----"
rg -n "budget_tokens|effort'\s*:|effort\b" packages/ai-anthropic/src -S || true
echo "---- Show lines around AnthropicThinkingOptions occurrences in model-meta.ts ----"
FILE="packages/ai-anthropic/src/model-meta.ts"
for L in 700 720 740 750 760 820 840 860 900 945 950; do
# best-effort; only print if within file
if [ "$(wc -l < "$FILE")" -ge "$L" ]; then
sed -n "$((L-5)),$((L+10))p" "$FILE" | awk -v start="$L" 'NR==1{printf "=== around %s ===\n", start} {print}'
fi
doneRepository: TanStack/ai
Length of output: 8295
🏁 Script executed:
#!/bin/bash
set -euo pipefail
FILE="packages/ai-anthropic/src/text/text-provider-options.ts"
# Show AnthropicThinkingOptions interface + nearby comments
sed -n '70,230p' "$FILE" | awk '{print}' | nl -ba | sed -n '1,200p'
echo
echo "---- Also show relevant adapter usage for thinking/effort ----"
ADAPTER="packages/ai-anthropic/src/adapters/text.ts"
sed -n '330,520p' "$ADAPTER" | awk '{print}' | nl -ba | head -n 220Repository: TanStack/ai
Length of output: 98
🏁 Script executed:
#!/bin/bash
set -euo pipefail
FILE="packages/ai-anthropic/src/text/text-provider-options.ts"
ADAPTER="packages/ai-anthropic/src/adapters/text.ts"
echo "== AnthropicThinkingOptions + comments (text-provider-options.ts:70-230) =="
awk '{printf "%5d\t%s\n", NR, $0}' <(sed -n '70,230p' "$FILE")
echo
echo "== adapter text.ts: thinking/effort handling (text.ts:330-520) =="
awk '{printf "%5d\t%s\n", NR, $0}' <(sed -n '330,520p' "$ADAPTER") | head -n 260Repository: TanStack/ai
Length of output: 15363
🏁 Script executed:
#!/bin/bash
set -euo pipefail
FILE="packages/ai-anthropic/src/model-meta.ts"
echo "== Search for AnthropicAdaptiveThinkingOptions and AnthropicEffortOptions in model-meta.ts =="
rg -n "AnthropicAdaptiveThinkingOptions|AnthropicEffortOptions" "$FILE" || true
echo
echo "== Search for 'extended_thinking' usage in model-meta.ts =="
rg -n "extended_thinking" "$FILE" || true
echo
echo "== Show imports near top of model-meta.ts (first 80 lines) =="
sed -n '1,120p' "$FILE" | awk '{print NR"\t"$0}'
echo
echo "== Show how Fable 5 typing is declared around type mapping near lines ~900-970 =="
sed -n '900,970p' "$FILE" | awk '{print NR"\t"$0}'Repository: TanStack/ai
Length of output: 7337
🏁 Script executed:
#!/bin/bash
set -euo pipefail
rg -n "validateTextProviderOptions|InternalTextProviderOptions|AnthropicTextProviderOptions" packages/ai-anthropic/src -S
echo "---- extended_thinking / CLAUDE_FABLE_5 usage ----"
rg -n "extended_thinking|CLAUDE_FABLE_5|claude-fable-5" packages/ai-anthropic/src -S
echo "---- show validateTextProviderOptions implementation (if any) ----"
rg -n "function validateTextProviderOptions|const validateTextProviderOptions|validateTextProviderOptions\s*=" packages/ai-anthropic/src -S
# Print candidate file sections around validateTextProviderOptions
FILE_MATCHES=$(rg -l "validateTextProviderOptions" packages/ai-anthropic/src -S || true)
for f in $FILE_MATCHES; do
echo
echo "== $f =="
sed -n '1,240p' "$f" | head -n 240
doneRepository: TanStack/ai
Length of output: 21944
🏁 Script executed:
#!/bin/bash
set -euo pipefail
FILE="packages/ai-anthropic/src/text/text-provider-options.ts"
sed -n '220,370p' "$FILE" | awk '{printf "%5d\t%s\n", NR+220, $0}'Repository: TanStack/ai
Length of output: 4262
🏁 Script executed:
#!/bin/bash
set -euo pipefail
FILE="packages/ai-anthropic/src/model-meta.ts"
echo "== ProviderOptions mapping for non-extended models around comments ~860-940 =="
sed -n '850,930p' "$FILE" | awk '{print}' | head -n 240
echo
echo "== ProviderOptions mapping around CLAUDE_FABLE_5 entry (~930-960) =="
sed -n '930,970p' "$FILE" | awk '{print}'
echo
echo "== Check models with extended_thinking:false and what provider options they map to =="
# Print lines mentioning extended_thinking:false and nearby following provider option mapping entries
rg -n "extended_thinking:\s*false" "$FILE"
echo
echo "== Show around extended_thinking:false blocks near first occurrences (~420-520) =="
sed -n '400,520p' "$FILE" | awk '{print}' | head -n 260
echo
echo "== Show around extended_thinking:false blocks near ~450-490 for Haiku/others =="
sed -n '440,510p' "$FILE" | awk '{print}'Repository: TanStack/ai
Length of output: 9999
🏁 Script executed:
#!/bin/bash
set -euo pipefail
rg -n "output_config\.effort|output_config.*effort|effort.*output_config" packages/ai-anthropic/src -S || true
rg -n "model.*FABLE|fable" packages/ai-anthropic/src/adapters/text.ts packages/ai-anthropic/src/model-meta.ts -S || true
rg -n "effort'\s*,|effort\W" packages/ai-anthropic/src/adapters/text.ts -S || trueRepository: TanStack/ai
Length of output: 644
🏁 Script executed:
#!/bin/bash
set -euo pipefail
rg -n "adaptive_thinking\s*:" packages/ai-anthropic/src/model-meta.tsRepository: TanStack/ai
Length of output: 121
🏁 Script executed:
#!/bin/bash
set -euo pipefail
FILE="packages/ai-anthropic/src/model-meta.ts"
echo "== Around line ~130-190 =="
sed -n '120,200p' "$FILE" | awk '{print NR "\t" $0}'
echo
echo "== Around line ~710-750 =="
sed -n '710,750p' "$FILE" | awk '{print NR "\t" $0}'Repository: TanStack/ai
Length of output: 3042
🏁 Script executed:
#!/bin/bash
set -euo pipefail
FILE="packages/ai-anthropic/src/model-meta.ts"
echo "== model-meta.ts: lines 710-770 =="
awk 'NR>=710 && NR<=770 {printf "%5d\t%s\n", NR, $0}' "$FILE"
echo
echo "== model-meta.ts: lines 930-970 =="
awk 'NR>=930 && NR<=970 {printf "%5d\t%s\n", NR, $0}' "$FILE"Repository: TanStack/ai
Length of output: 3910
Fix AnthropicThinkingOptions typing for CLAUDE_FABLE_5
packages/ai-anthropic/src/model-meta.ts sets CLAUDE_FABLE_5.supports.extended_thinking: false but the model’s provider options still include AnthropicThinkingOptions (lines 722-760, 945-952). Claude Fable 5 only supports “adaptive thinking” via effort and does not accept thinking: { type: 'enabled', budget_tokens } (400). Remove AnthropicThinkingOptions from the Fable 5 provider-options type and type the adaptive/effort controls instead.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@packages/ai-anthropic/src/model-meta.ts` around lines 722 - 760,
CLAUDE_FABLE_5 is typed with AnthropicThinkingOptions but the model only
supports adaptive/effort controls (extended_thinking: false); remove
AnthropicThinkingOptions from the generic type list for CLAUDE_FABLE_5 and
replace it with the specific adaptive/effort options type used by your codebase
(e.g., AnthropicAdaptiveThinkingOptions or AnthropicEffortOptions) so the
provider-options reflect effort/`effort`-style controls instead of the
`thinking: { type: 'enabled', budget_tokens }` shape; update the satisfies
ModelMeta<...> clause for CLAUDE_FABLE_5 accordingly and ensure imports/types
are adjusted to reference the adaptive/effort options symbol.
Comment on lines
+39
to
+44
| const response = await fetch(url, { | ||
| headers: { | ||
| 'x-api-key': apiKey, | ||
| 'anthropic-version': '2023-06-01', | ||
| }, | ||
| }) |
Contributor
There was a problem hiding this comment.
Add a timeout/abort to the Anthropic API request.
At Line 39, fetch has no timeout, so a stalled connection can hang the sync job indefinitely. Use AbortSignal.timeout(...) (or an AbortController) and fail fast.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@scripts/fetch-anthropic-models.ts` around lines 39 - 44, The fetch call in
scripts/fetch-anthropic-models.ts (the const response = await fetch(url, { ...
}) line) needs an abort timeout to avoid hanging; create an AbortController (or
use AbortSignal.timeout) and pass its signal in the fetch options, set a
sensible timeout (e.g., 10s) and ensure you handle the abort by catching the
thrown error and logging/propagating it back from the function that calls fetch;
keep the existing headers ('x-api-key' and 'anthropic-version') and ensure any
manual timer is cleared on success when using AbortController.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
🎯 Changes
Adds Anthropic's Claude Fable 5 model, moves Anthropic model syncing onto the first-party Models API, and fixes incorrect Anthropic model metadata uncovered along the way.
New model: Claude Fable 5
claude-fable-5on the Anthropic adapter andanthropic/claude-fable-5on the OpenRouter adapter: model metadata, per-model provider option types, input modality types, and server-tool capability types.ANTHROPIC_MODELShas entries in all three per-model type maps.Anthropic sync from the first-party Models API
scripts/fetch-anthropic-models.tssnapshotsGET /v1/modelstoscripts/anthropic.models.json(committed snapshot is genuine API output, 11 models).scripts/sync-anthropic-models.tsadds new catalog models topackages/ai-anthropic/src/model-meta.ts— constants,ANTHROPIC_MODELS, and all three type maps — and writes a changeset. Pricing isn't exposed by the API, so a hand-maintainedPRICINGtable is the source of truth; a new model fails the sync loudly until a row is added.@tanstack/ai-anthropic(OpenRouter's dotted slugs likeclaude-opus-4.8are not valid Anthropic API ids).sync-models.ymlworkflow now runs the Anthropic sync too:pnpm generate:modelschains the Anthropic fetch + sync after the OpenRouter steps. Requires a newANTHROPIC_API_KEYrepo secret before merge — without it the nightly job fails (intentionally loud, but it also blocks the OpenRouter half).Metadata corrections (verified against the live Models API and pricing docs)
claude-opus-4-5is $5/MTok in, $25/MTok out (was wrongly $15/$75); cache-read (cached) prices added for all models that publish them.claude-opus-4.8/claude-opus-4.8-fastrenamed to the real API idsclaude-opus-4-8/claude-opus-4-8-fast(the dotted ids 404 against the Anthropic API; three docs pages fixed too). Note: this changes a published type-map key, but the dotted id never worked at runtime.claude-fable-5,claude-opus-4-7-fast,claude-opus-4-8, andclaude-opus-4-8-fastwere missing fromAnthropicChatModelToolCapabilitiesByName, so the typed API rejected every server tool on those models. Fixed, and the sync script now maintains that map.@anthropic-ai/sdkbumped to ^0.104.0 (itsModelunion includesclaude-fable-5andclaude-opus-4-8).No E2E changes: the E2E matrix is provider-granular with pinned model ids, and this change has no runtime behavior — model metadata is consumed as types; type tests run under
test:lib/test:types.✅ Checklist
pnpm run test:pr.🚀 Release Impact
🤖 Generated with Claude Code
Summary by CodeRabbit
New Features
Bug Fixes
Documentation