Claude Fable 5.2 API for Coding and Long-Horizon Agents

Claude Fable 5.2 is Anthropic's highest-capability generally available model for open-ended technical work. Released on November 9, 2026, it is designed for tasks that may run for hours or days: software migrations, agentic coding, deep research, scientific analysis, and projects that require repeated planning, tool use, verification, and revision.

The model accepts text and images and returns text. It has a 1,000,000-token context window, supports up to 128,000 output tokens, and uses adaptive thinking on every request. Developers control reasoning depth with the effort setting rather than turning thinking off. Anthropic lists high as the API default.

GPTProto provides access to Fable 5.2 alongside the wider Claude model catalog. One GPTProto key and one balance can be used to test Fable against lower-cost models, route different stages of an agent to different providers, and keep billing in one account.

Claude Fable 5.2 API Applications

Repository-scale coding and migration: Give the model architecture notes, source files, tests, logs, and acceptance criteria, then let an agent inspect dependencies, implement changes, run tools, and revise failed steps. This is the clearest fit for Fable 5.2 in coding workflows.

Long-horizon agentic work: Use higher effort for loops that must plan, execute, inspect results, recover from errors, and continue toward a measurable outcome. Lower effort remains available for easier turns inside the same workflow.

Deep technical research: Combine a large evidence set with tool results and ask the model to maintain a research plan, compare competing explanations, verify calculations, and produce a finished report rather than a short summary.

Visual and document analysis: Pair screenshots, charts, diagrams, or scanned pages with text instructions. The model can reason over image input, but it returns text rather than generating images.

Advisor and orchestrator roles: Reserve Fable for the hardest decisions while smaller models handle routine subtasks. This reduces the need to run the highest-cost model for every turn in a multistep system.

How Claude Fable 5.2 Compares with Other Frontier Models

The useful questions are not simply which model is newest. Claude Fable 5.2 vs Fable 5.1 is mainly an upgrade decision; Claude Fable 5.2 vs Opus 5.5 is a cost-versus-task-difficulty decision; GPT-6 Astra vs Claude Fable 5.2 is often a choice between two agent stacks and their tool behavior.

Anthropic reported the following results at each model's stated evaluation setting. They were not independently reproduced by GPTProto, so use them to choose candidates for your own evals rather than as a guarantee.

FrontierCode scores for Fable 5.2 and Fable 5.1 use xhigh effort; Anthropic labels the Gemini 4 Carbon score as medium effort. The launch comparison names Gemini 4 Carbon, not Gemini 4 Argon. Do not reuse Carbon's scores for Gemini 4 Argon; compare the available models on the same prompts in the GPTProto playground.

Claude Fable 5.2 API Migration Checklist

Moving from Claude Fable 5.1 is mostly a model upgrade, but four changes can break an existing agent:

1. Update the model route. Anthropic's official ID changes from claude-fable-5-1 to claude-fable-5-2. For GPTProto, use the exact model string displayed in this page's API Usage tab.

2. Keep the conversation prefix unchanged. Replaying a thinking block after editing the earlier system prompt, tools, or messages can return a 400 prefix-mismatch error. Append new turns instead of rewriting prior ones.

3. Check cross-model fallbacks. Fable 5.2 and Mythos 5.2 can read Fable 5.2 thinking blocks; other models may drop them. Store final task state outside hidden reasoning before switching models.

4. Retest tools and reasoning controls. Adaptive thinking stays on. Native Anthropic requests cannot force a specific tool with tool_choice: any or tool; older computer-use declarations may also require an update. Re-run tool, latency, token, and refusal tests before moving production traffic.

When migrating from an official provider endpoint to GPTProto, also change the base URL and credential while preserving your application-level tests. Do not assume every provider-specific beta field is supported through an OpenAI-compatible route; follow the live API Usage example for the supported request shape.

When Should You Choose Claude Fable 5.2?

Choose Fable 5.2 when failure on a difficult, long-running task costs more than the additional tokens: large migrations, multi-hour coding agents, technical research, complex visual reasoning, or workflows that must recover and keep working after tool failures.

Start with Claude Opus 5.5 for most everyday coding and knowledge work, as Anthropic recommends. Use GPT-6 Astra when your evaluation favors OpenAI's agent and computer-use stack. Move up to Fable only when the same task and harness show a meaningful gain.