Claude Fable 5.2 Is Now on Felo OpenAPI: A Developer’s Guide
Explore Claude Fable 5.2 on Felo OpenAPI: capabilities, benchmarks, coding and research use cases, pricing context, and how to get started.
Claude Fable 5.2 Is Now on Felo OpenAPI: A Developer's Guide
Claude Fable 5.2 is built for the kind of work that rarely fits into a single prompt: migrating a large codebase, investigating a difficult research question, or coordinating a long sequence of tool calls. Following Anthropic's November 9, 2026 launch, developers can explore the model through Felo OpenAPI.
This guide covers what Fable 5.2 can do, how it compares with its predecessor in Anthropic's published evaluations, and how to start planning an integration. We'll also look at where a smaller model may be a better fit.
alt="Claude Fable 5.2 on Felo OpenAPI: 1M context, 128K maximum output, and adaptive thinking" src="https://blog-static.felo.ai/claude_fable_5_2_infographic_cover_3bc88b5669.png" />
What is Claude Fable 5.2?
Claude Fable 5.2 is Anthropic's model for demanding reasoning and long-horizon agentic work. Anthropic positions it for ambitious technical problems where a model needs to work through multiple steps, use tools, and carry context across a substantial task.
That might mean examining a repository before proposing a migration, reviewing charts alongside written evidence, or turning an open-ended research assignment into a structured deliverable. The model can accept text and images and generate text; tools and application integrations provide the surrounding environment for actions such as running code or retrieving documents.
Anthropic also released Claude Mythos 5.2 alongside it. The two share the same underlying model, but Fable includes cybersecurity and biology safeguards, while access to Mythos is limited to organizations verified for eligible uses. Felo's model discussed here is Claude Fable 5.2. Source: Anthropic launch announcement.
Model specifications at a glance
| Specification | Anthropic's published specification |
|---|---|
| Release date | November 9, 2026 |
| Context window | 1 million tokens |
| Maximum output | 128,000 tokens |
| Input and output | Text and images in; text out |
| Thinking | Adaptive, always on |
| Default effort in Claude API | High |
| Knowledge cutoff | June 2026 |
| Claude API model ID | claude-fable-5-2 |
These are Anthropic's model specifications. Available context, parameters, routes, and identifiers can depend on the integration. Check the Felo model page for the current Felo configuration. Source: Anthropic model documentation.
What is new in Claude Fable 5.2?
Stronger coding and longer-running work
Fable 5.2 is intended for projects that require understanding more than one file or producing more than one patch. Anthropic reports improvements in coding and reasoning, including prerelease tests in which the model helped its engineers complete difficult software migrations.
For a development team, that makes a repository-wide dependency upgrade a useful test case. An agent could inspect the codebase, identify affected interfaces, propose a migration sequence, implement changes, and run an existing test suite. The important word is could: the result depends on the tools and permissions supplied by the application, the quality of the tests, and human review.
Start with a bounded task. Give the agent a clear definition of success, the commands it is allowed to run, and a requirement to report failing tests and unresolved assumptions. A longer context window helps it consider more material, but it does not guarantee that every detail will be recalled or interpreted correctly.
Adaptive thinking and configurable effort
Anthropic says adaptive thinking is always enabled for Fable 5.2 on the Claude API. Developers can use an effort setting to influence how much reasoning the model applies. Higher effort may help on harder problems, while a lower setting may be more practical for a task with a tighter latency or cost budget.
The right setting is empirical. Try the same representative tasks at different supported effort levels, then compare correctness, completion time, token usage, and total cost per successful task. Verify which controls your Felo route exposes before adding them to a production request.
Anthropic also says Fable 5.2 can produce output up to 30% faster than Fable 5.1, and offers an optional fast mode on its own platform. Those are Anthropic's claims and product settings; Felo route availability and performance can differ. Source: Anthropic announcement.
A 1-million-token context window
A large context window gives developers room to bring more of a problem into one request: specifications, relevant source files, logs, and prior decisions, for example. It can also support analysis across a collection of longer documents.
More context is not automatically better. Supplying irrelevant files increases the material the model has to process and can increase cost. Select the most useful sources, label them clearly, and ask the model to cite the files or passages behind important conclusions. Keep a separate, durable record of decisions when a workflow spans multiple sessions.
Better vision for charts and technical material
Fable 5.2 accepts image input, which can help when useful information appears in a diagram, screenshot, or chart rather than plain text. For example, a team could ask it to compare an architecture diagram with a written system specification and identify apparent inconsistencies.
Image interpretation should still be checked against the source, especially for small labels, precise figures, and safety-critical decisions. Image and document handling may vary by provider and API route, so confirm supported input formats on the Felo model page before implementation.
Claude Fable 5.2 benchmarks: what Anthropic measured
Anthropic published the following results for Fable 5.2 and Fable 5.1 in its launch announcement:
| Evaluation | What it measures | Fable 5.2 | Fable 5.1 |
|---|---|---|---|
| FrontierCode 2.0 | Agentic coding | 49.1% | 32.4% |
| FrontierSWE v2 | Agentic software engineering | 69.9% | 56.3% |
| AutomationBench | Business workflows | 51.4% | 31.4% |
| Terminal-Bench-Science 0.1 | Agentic scientific research | 74.6% | 52.6% |
| Chartography | Visual chart recognition | 73.4% | 44.8% |
The FrontierCode figures use xhigh effort. Anthropic says scores generally use the highest available effort unless otherwise noted, with safeguards on for these Fable results. The company also publishes benchmark-specific details in its announcement and linked system card.
The numbers show improvement on these particular tests; they are not a promise of the same improvement on your workload. Benchmark results depend on prompts, tools, scoring, and test conditions. Run your own evaluation using tasks your team actually cares about, including difficult edge cases and examples where a wrong answer is costly.
Where can developers use Claude Fable 5.2?
1. Repository-wide coding tasks
Use Fable 5.2 to investigate issues that cross module boundaries: an API migration, a dependency upgrade, or a bug involving multiple services. Provide a repository map, the relevant requirements, and a way to run tests. Ask for a plan before granting write access, and review the final diff as you would a teammate's.
A useful prompt might be:
Inspect the authentication flow and propose a migration to the new session interface. Identify affected files and compatibility risks. Implement the smallest coherent change, run the relevant tests, and report what remains unverified.
This prompt defines a deliverable and a verification step. It does not assume the model has access to a repository or terminal: your agent application must supply those tools.
2. Technical research with a traceable deliverable
For a literature review or architecture decision, give the model a question, a curated set of sources, and the intended audience. Ask it to separate evidence from inference and flag conflicting claims. If your application provides search or retrieval tools, define which sources it may use and require links back to the original material.
One practical output is a decision memo with alternatives, trade-offs, open questions, and a recommendation. A human should verify the citations and any facts that drive an important decision. The June 2026 knowledge cutoff also means that recent information requires current sources.
3. Analysis of visual and written evidence
An operations team might compare a dashboard screenshot with an incident report; a product team might compare interface screenshots with acceptance criteria. Give the model the image and the relevant written context, then ask it to list observed evidence separately from its interpretation.
Where an exact number matters, check the original data. A readable chart is useful input, but it is not a substitute for a validated dataset.
4. Advisor and orchestration workflows
Anthropic describes Fable 5.2 as suitable for an advisor role, where another model consults it on a difficult problem, or an orchestrator role, where it delegates simpler tasks to smaller models. For example, a smaller model might classify incoming requests, while Fable reviews the subset requiring complex reasoning.
Your application must implement the routing, tool access, and handoffs. Measure the whole workflow: a more expensive model can be worthwhile on the hardest cases, but unnecessary escalation can raise cost and latency without improving the outcome.
How to get started with Claude Fable 5.2 on Felo OpenAPI
Felo OpenAPI brings model access and developer integrations into one platform. To get started:
- Review the model page. Open the Claude Fable 5.2 page and check the currently listed model ID, supported protocols, providers, and prices.
- Create a Felo API key. Follow the getting-started guide and store the key in an environment variable or secret manager. Felo's guide says model API use requires service credits.
- Choose your client and protocol. For Claude Code and other Anthropic-compatible clients, Felo documents
https://openapi.felo.ai/apias the base URL. See the Claude Code setup guide for configuration details. - Select the published Felo model ID. Copy it directly from the model page or catalog. Anthropic's own ID is
claude-fable-5-2; confirm the ID and route to use on Felo rather than assuming they are identical. - Run a small test. Send a short request first, then test a representative task with your actual input size, tool configuration, and evaluation criteria. Check usage and errors before scaling up.
For a coding agent, keep API keys out of source control. Give the agent only the repository and tool permissions it needs, and require review before changes reach production.
Pricing, limits, and operational considerations
Anthropic's published standard Claude API rates are $8 per million input tokens and $40 per million output tokens. Anthropic also lists separate rates for caching and fast mode. These are Anthropic's prices, not a quote for Felo: consult the Felo model page for current provider-specific rates and conditions. Source: Anthropic model documentation.
Token prices alone do not tell you the cost of a completed job. A long agent run may include repeated requests, tool results, retries, and substantial context. Set usage limits, log token consumption, and compare cost per accepted result across model and effort settings.
Anthropic documents model-specific restrictions for its Claude API, including always-on adaptive thinking, restrictions on certain sampling and forced-tool-choice settings, and a 30-day data-retention requirement unless otherwise authorized. Confirm the applicable controls, retention terms, and provider policies for your chosen Felo route before sending sensitive data. Source: Anthropic model documentation.
When should you choose Fable 5.2?
Choose it for work where difficult reasoning, a large body of context, or sustained multi-step execution matters enough to justify testing a more capable model: a complex migration, a demanding research workflow, or an agent that must investigate before acting.
For routine classification, short extraction, or high-volume simple requests, test a smaller model first. Anthropic itself recommends starting with Claude Opus 5.5 for most coding and knowledge tasks, then moving to Fable when the task demands it. Your own evaluation should determine the right balance of quality, speed, and cost. Source: Anthropic announcement.
Frequently asked questions
Is Claude Fable 5.2 available through Felo OpenAPI?
Felo OpenAPI supports Claude Fable 5.2. Visit the model page for the latest routing, availability, and integration details.
What is the context window of Claude Fable 5.2?
Anthropic lists a 1-million-token context window and a maximum output of 128,000 tokens. Check the limits of your selected Felo provider and route before building around those maximums.
Can Claude Fable 5.2 understand images?
Yes. Anthropic documents text and image input with text output. Supported input formats and request syntax depend on the API and route you use.
Is Claude Fable 5.2 the same as Claude Mythos 5.2?
They share an underlying model, according to Anthropic, but differ in safeguards and access. Mythos is available to eligible verified organizations; Fable has safeguards for cybersecurity and biology uses.
Does Claude Fable 5.2 browse the web or execute code by itself?
A model can reason about information and request tool use, but browsing and code execution require tools supplied by the application or platform. Configure those tools and permissions explicitly, and verify what each provider supports.
For your next difficult coding or research workflow, start with a small, measurable task and see how Claude Fable 5.2 performs with your own context and evaluation criteria.