Claude Fable 5.2
For demanding reasoning and long-horizon agentic work
Modalities
In / Out Price
10% off$7.2/ 1M
$8 Official price
$36/ 1M
$40 Official price
Context
1M
Released
Nov 9, 2026
API call examples
POST /v1/chat/completionscurl --request POST 'https://api.tokatlas.com/v1/chat/completions' \
--header "Authorization: Bearer $UNIROUTER_API_KEY" \
--header 'Content-Type: application/json' \
--data '{
"model": "claude-fable-5-2",
"messages": [
{
"role": "user",
"content": "Explain quantum entanglement."
}
],
"temperature": 0.7
}'Providers & routes
Current provider · Anthropic
Platform configured price · Group: ccmax
USD / 1M tokens. Each row shows input and output from the same tier. Displayed rates use the lowest available group multiplier; your API key group may differ.
Claude Fable 5.2 for sustained technical work
Use it to break work with several source materials into reviewable steps, such as tracing dependencies between modules or identifying conclusions that need more evidence.
Long context
Keep design specs, tests, and approved decisions in one working context.
Adaptive reasoning
Set effort for the task and ask for evidence behind decisions that need review.
Phased verification
Break migrations into verifiable steps and update the next phase with real test results.
Visual and tool evidence
Bring screenshots and tool results together and ask for evidence behind each conclusion.
About
Long-running reasoning and output
Fable 5.2 can keep specs, code, and repeated verification in a single working context. The provider lists a 1,000,000-token window and 128,000 output tokens, with text and image input. Adaptive reasoning stays enabled and the API defaults to high effort.
Work that needs repeated revision
Use it for migrations, technical research, and work spanning multiple documents. Keep approved decisions, tool logs, and acceptance criteria distinct so a resumed task can identify the verified information it relies on.
Choosing among Fable and Opus
Compare candidates on the same inputs. For extended tasks, record rework, discovery of dependencies, and acceptance-test completion. Benchmark short jobs separately from extended work.
Migration checks
After updating the model ID, check effort, tool result formats, and output budgets. Reuse previous failure cases to verify recovery, completion, and the evidence left behind by the run.
Research and multi-document delivery need different checks. Research findings need sources and unresolved points. Migration documents need dependencies, sequencing, and rollback criteria. For longer assignments, leave an independently reviewable artifact at each checkpoint.
Claude Fable 5.2 FAQ
How much room is available for a long document and its response?
Fable 5.2 has a one-million-token context and a 128K generation ceiling. Leave room for reasoning and the answer when assembling a large input.
When is it worth comparing Fable 5.2 with Opus 5.5?
Compare them on ambiguous work that requires repeated planning. Keep tools and acceptance criteria constant and measure the cost of finishing the whole task.
What should I retest when upgrading from Fable 5.1?
Pick a recurring research or development task and hold its prompt, references and tools constant. Compare output speed and the manual corrections each version needs.
What should I record when comparing Fable with Gemini 4?
Record the exact variant. Results for Gemini 4 Carbon should not stand in for Argon when deciding which model to use.