Claude Fable 5.2 API
Call Claude Fable 5.2 with your KeelRouter API key, the same key you use for every other model. From $2.50 / $12 per 1M tokens, paid per call from a prepaid balance. Failed calls are free.
What is new
- Released by Anthropic on 2026-11-10. Leads Claude Opus 5.5 on coding, autonomous agent loops, headless 3D/SVG/Canvas generation and high-precision code review.
- Generates ~30% fewer output tokens per turn than Opus 5.5 at the same effort level; add an explicit completion rule when you want full-pass execution without early check-ins.
- Supports effort levels (low, medium, high, xhigh, max), 1M context window and 128K max output tokens on /v1/chat/completions and /v1/messages.
- Omit temperature, top_p and top_k: any non-default sampling value returns HTTP 400.
Pricing
| Option | Price |
|---|---|
| Input | $2.50 per 1M tokens |
| Output | $12 per 1M tokens |
Token cost calculator (Claude Fable 5.2)
$0.37
Save 69% per call
Compared with OpenRouter
Options where KeelRouter costs less for Claude Fable 5.2.
| Option | KeelRouter | OpenRouter |
|---|---|---|
| Input | $2.50 per 1M tokens | $8 per 1M tokens |
| Output | $12 per 1M tokens | $40 per 1M tokens |
OpenRouter list prices from its public model catalog, checked 2026-11-12; long-context surcharges not shown.
Call it
OpenAI-compatible endpoint /v1/chat/completions. Anthropic SDKs work on /v1/messages.
curl https://api.keelrouter.com/v1/chat/completions \
-H "Authorization: Bearer $KEELROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-fable-5-2",
"messages": [
{
"role": "user",
"content": "Write a haiku about the sea."
}
]
}'Prompt templates
Copy-ready prompts tuned for Claude Fable 5.2Autonomous coding agent (stops early exits)
effort: high | xhighFable 5.2 uses ~30% fewer tokens per turn than Opus 5.5 and checks in early on ambiguous tasks. This system prompt enforces full end-to-end execution.
You are an autonomous software engineering agent. Execution rules: 1. Do not stop to ask for confirmation or summarize partial progress until the entire task is complete and verified. 2. Read the surrounding codebase before editing. Fix root causes rather than adding surface-level fallbacks or speculative try/catch wrappers. 3. Do not modify files, exports, or tests outside the scope of the request. 4. Before finishing, run the relevant typecheck, linter, and unit tests, and fix any failures caused by your changes.
High-precision code review (zero false positives)
effort: highReproduces the 4/4 precision benchmark pattern from community evaluations: reports only real state/logic/security defects.
Review the git diff below for production defects. Rules: - Report only concrete bugs that will cause incorrect runtime behavior, data loss, race conditions, or security vulnerabilities. - Do not report style preferences, naming opinions, or hypothetical edge cases that cannot occur given the caller types. - For each finding, provide: (1) file and line range, (2) exact failure scenario and trigger input, and (3) minimal patch.
Deterministic Three.js scene with headless frame hook
effort: xhighExposes window.renderAt(seconds) so headless Chromium can scrub and verify frames at t = 0, 2, 5s.
Build a single-file HTML Three.js scene of a mechanical brass Orrery with orbiting planets, engraved rings, and warm area lighting. Requirements: 1. Procedural geometry and procedural PBR materials only (no external textures or model URLs). 2. Expose window.renderAt(t) that deterministically sets all orbital angles and camera position for timestamp t (in seconds) and renders one frame. 3. Include an interactive orbit camera and play/pause scrubber when opened in a normal browser.
Single-file playable browser game + automated self-test
effort: xhighAdds window.debugState() so Playwright can verify collision and scoring automatically.
Create a single-file HTML5 Canvas top-down wave survival game ("Void Drift") with crisp vector ship rendering, directional thruster particles, enemy flocking, and upgrade cards between waves.
Requirements:
1. All rendering, physics, and WebAudio synthesized sound effects in one self-contained HTML file.
2. Maintain 60 FPS with delta-time stepping.
3. Expose window.debugState() returning { score, wave, playerHp, activeEnemies, gameOver } for automated browser verification.Parameters
| Name | Type | Notes |
|---|---|---|
| model | string | Required. claude-fable-5-2 |
| messages | array | Required. OpenAI chat format |
| max_tokens | integer | Output cap; also sets the hold |
| stream | boolean | Server-sent events; usage is billed at the end |
| temperature | number | Not supported: any value other than the default returns a 400 error, and so do top_p and top_k. Omit them |
| tools | array | Passed through when the model supports them |
FAQ
- How much does the Claude Fable 5.2 API cost on KeelRouter?
- From $2.50 / $12 per 1M tokens. You pay per call from a prepaid USD balance; failed calls are not charged.
- Is Claude Fable 5.2 cheaper on KeelRouter than on OpenRouter?
- For these options, yes: Input $2.50 vs $8 per 1M tokens; Output $12 vs $40 per 1M tokens. OpenRouter list prices from its public model catalog, checked 2026-11-12; long-context surcharges not shown. No subscription; top up from $5 and pay per call.
- How do I call Claude Fable 5.2?
- Send an OpenAI-compatible request to /v1/chat/completions with model "claude-fable-5-2" and your KeelRouter API key.
- Is this the official Claude Fable 5.2 API?
- No. KeelRouter is an independent API gateway. Requests run on Claude Fable 5.2 through our upstream providers, and the language model keeps its own safety filters.
- How long are results kept?
- We keep request records (tokens and cost), not your prompts or answers.
More from Anthropic
KeelRouter is an independent service and is not affiliated with Anthropic. Claude Fable 5.2 is a trademark of its owner, used here only to name the model a request runs on.