Technical Specifications of Claude Fable 5.1
| Specification | Claude Fable 5.1 |
|---|---|
| Provider | Anthropic |
| Model ID | claude-fable-5-1 |
| Model family | Claude Fable 5.1 / Mythos-class |
| Release date | September 1, 2026 |
| Context window | 1,000,000 tokens |
| Maximum output | 128,000 tokens |
| Input | Text, images |
| Output | Text |
| Thinking | Adaptive thinking, always on |
| Default effort | high |
| Effort levels | low, medium, high, xhigh, max |
| Knowledge cutoff | June 2026 |
| API platforms | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
| Comparative latency | Slower |
| Status | Active |
Anthropic positions Claude Fable 5.1 specifically for demanding reasoning and long-horizon agentic work. Its 1M-token context and 128K maximum output make it suitable for large repositories, lengthy research material, and extended multi-step workflows.
What Is Claude Fable 5.1?
Claude Fable 5.1 is Anthropic's September 2026 upgrade to the Fable 5 model, targeting complex reasoning, long-running agentic coding, multi-step research, and professional knowledge work.
Rather than simply increasing the model's general-purpose intelligence, Fable 5.1 focuses heavily on task completion over long horizons. Anthropic specifically highlights improvements in agentic coding, research, and document, spreadsheet, and presentation workflows.
Fable 5.1 belongs to Anthropic's Mythos-class tier. Claude Mythos 5.1 shares the same underlying capabilities but is restricted to Project Glasswing participants, while Fable 5.1 is the generally available safeguarded version.
Main Features of Claude Fable 5.1
- 1M-token long context: The model can maintain extremely large amounts of source material, repository context, conversation history, or research evidence within a single context window.
- Long-horizon agentic execution: Fable 5.1 is optimized for tasks requiring many sequential actions rather than isolated question-answering, including extended coding and research workflows.
- Adaptive thinking: Thinking is always enabled, while the
effortsetting controls the amount of reasoning allocated to a request. Available levels arelow,medium,high,xhigh, andmax. - Strong software engineering: Anthropic reports improvements on agentic coding evaluations, including Terminal-Bench 4.0, CursorBench, and AutomationBench.
- Multimodal document understanding: Fable 5.1 accepts text and image inputs, making it suitable for documents, screenshots, diagrams, spreadsheets, and other visual material.
- Lower cache-read cost: Prompt-cache reads are priced at $0.25 per million tokens, 75% below Fable 5's $1 per million cached tokens. Anthropic estimates this can reduce typical workload costs by about 25% and highly agentic workloads by up to approximately 45%.
Benchmark Performance of Claude Fable 5.1
Anthropic's September 1 evaluation compares Fable 5.1 with Fable 5, Claude Opus 5, and GPT-5.6 Sol. The strongest gains appear on long-running agentic tasks rather than conventional knowledge benchmarks.
| Benchmark | Fable 5.1 | Fable 5 | Opus 5 | GPT-5.6 Sol |
|---|---|---|---|---|
| Terminal-Bench-Science 0.1 | 52.6% | 24.7% | 29.0% | 22.4% |
| Terminal-Bench 4.0 | 55.8% | 42.0% | 52.3% | 37.3% |
| AutomationBench | 31.4% | 17.1% | 26.9% | 19.6% |
| CursorBench 3.2.0 | 73.4% | 70.5% | 70.0% | 67.2% |
| OSWorld 2.0 — partial | 77.9% | 72.9% | 75.4% | — |
| OSWorld 2.0 — strict | 41.7% | 36.1% | 39.6% | — |
| Humanity's Last Exam — no tools | 60.9% | 57.8% | 56.6% | — |
| Humanity's Last Exam — with tools | 65.0% | 63.8% | 63.6% | — |
| GDPval-AA v2 | 1,853 Elo | 1,723 | 1,824 | 1,711 |
The largest improvement is on Terminal-Bench-Science 0.1, where Fable 5.1 reaches 52.6%, compared with 24.7% for Fable 5. On GDPval-AA v2, Fable 5.1 reaches 1,853 Elo, ahead of Opus 5 at 1,824 and Fable 5 at 1,723.
Artificial Analysis independently reports that Fable 5.1 reaches a 66 Intelligence Index score at maximum effort, placing it ahead of Opus 5 at 63 and Fable 5 at 60 in that evaluation.
Benchmark results should nevertheless be interpreted by workload. For example, third-party benchmark aggregation reports Fable 5.1 at 89.1% on SWE-bench Multilingual versus 89.5% for Opus 5, illustrating that Fable 5.1 does not automatically dominate every specialized evaluation.
Claude Fable 5.1 vs Claude Fable 5 vs Claude Opus 5
| Model | Context | Max Output | Primary Position | Default Effort |
|---|---|---|---|---|
| Claude Fable 5.1 | 1M | 128K | Long-horizon reasoning and agentic work | High |
| Claude Fable 5 | 1M+ | 128K | Previous Fable generation | High |
| Claude Opus 5 | 1M | 128K | Complex agentic coding and enterprise work | High |
Fable 5.1 is the stronger choice when the workload involves long-running autonomous execution, demanding reasoning, multi-step research, or complex coding. Anthropic itself recommends starting most workloads with Opus 5 and moving to Fable 5.1 when higher-effort Opus 5 evaluations remain insufficient.
Compared with Fable 5, the major changes are improved agentic performance, lower cache-read costs, per-message effort controls, progress updates, and changes to thinking-block and tool-use behavior.
Limitations of Claude Fable 5.1
Fable 5.1 is not the fastest or cheapest Claude model. Anthropic classifies its comparative latency as slower, and its standard input/output rates are higher than Opus 5 and Sonnet 5.
It also has important API migration differences from Fable 5. In particular, forced tool use with tool_choice: any or tool returns a 400 error, while auto and none remain supported. Applications migrating from Fable 5 should also account for changes involving thinking blocks and conversation history.
Fable 5.1 also operates with production safety classifiers. Some restricted cybersecurity and biology requests can trigger refusal or fallback behavior, so benchmark results on tasks affected by safeguards should not be interpreted as unrestricted Mythos-class capability.
Best Use Cases for Claude Fable 5.1
- Repository-scale software engineering — multi-file refactoring, debugging, migration, testing, and long-running coding agents.
- Autonomous research — multi-step investigation requiring search, evidence collection, analysis, and synthesis.
- Large-document analysis — processing extensive technical documentation, contracts, reports, or research collections.
- Computer-use workflows — tasks involving desktop interfaces, spreadsheets, presentations, and other visual environments.
- Enterprise knowledge work — complex reports, financial or analytical workflows, and professional deliverables requiring sustained reasoning.
- Agentic systems — applications where the model must repeatedly call tools, inspect results, revise its plan, and continue until a task is complete.Technical Specifications of Claude Fable 5.1
| Specification | Claude Fable 5.1 |
|---|---|
| Provider | Anthropic |
| Model ID | claude-fable-5-1 |
| Model family | Claude Fable 5.1 / Mythos-class |
| Release date | September 1, 2026 |
| Context window | 1,000,000 tokens |
| Maximum output | 128,000 tokens |
| Input | Text, images |
| Output | Text |
| Thinking | Adaptive thinking, always on |
| Default effort | high |
| Effort levels | low, medium, high, xhigh, max |
| Knowledge cutoff | June 2026 |
| API platforms | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
| Comparative latency | Slower |
| Status | Active |
Anthropic positions Claude Fable 5.1 specifically for demanding reasoning and long-horizon agentic work. Its 1M-token context and 128K maximum output make it suitable for large repositories, lengthy research material, and extended multi-step workflows.
What Is Claude Fable 5.1?
Claude Fable 5.1 is Anthropic's September 2026 upgrade to the Fable 5 model, targeting complex reasoning, long-running agentic coding, multi-step research, and professional knowledge work.
Rather than simply increasing the model's general-purpose intelligence, Fable 5.1 focuses heavily on task completion over long horizons. Anthropic specifically highlights improvements in agentic coding, research, and document, spreadsheet, and presentation workflows.
Fable 5.1 belongs to Anthropic's Mythos-class tier. Claude Mythos 5.1 shares the same underlying capabilities but is restricted to Project Glasswing participants, while Fable 5.1 is the generally available safeguarded version.
Main Features of Claude Fable 5.1
- 1M-token long context: The model can maintain extremely large amounts of source material, repository context, conversation history, or research evidence within a single context window.
- Long-horizon agentic execution: Fable 5.1 is optimized for tasks requiring many sequential actions rather than isolated question-answering, including extended coding and research workflows.
- Adaptive thinking: Thinking is always enabled, while the
effortsetting controls the amount of reasoning allocated to a request. Available levels arelow,medium,high,xhigh, andmax. - Strong software engineering: Anthropic reports improvements on agentic coding evaluations, including Terminal-Bench 4.0, CursorBench, and AutomationBench.
- Multimodal document understanding: Fable 5.1 accepts text and image inputs, making it suitable for documents, screenshots, diagrams, spreadsheets, and other visual material.
- Lower cache-read cost: Prompt-cache reads are priced at $0.25 per million tokens, 75% below Fable 5's $1 per million cached tokens. Anthropic estimates this can reduce typical workload costs by about 25% and highly agentic workloads by up to approximately 45%.
Benchmark Performance of Claude Fable 5.1
Anthropic's September 1 evaluation compares Fable 5.1 with Fable 5, Claude Opus 5, and GPT-5.6 Sol. The strongest gains appear on long-running agentic tasks rather than conventional knowledge benchmarks.
| Benchmark | Fable 5.1 | Fable 5 | Opus 5 | GPT-5.6 Sol |
|---|---|---|---|---|
| Terminal-Bench-Science 0.1 | 52.6% | 24.7% | 29.0% | 22.4% |
| Terminal-Bench 4.0 | 55.8% | 42.0% | 52.3% | 37.3% |
| AutomationBench | 31.4% | 17.1% | 26.9% | 19.6% |
| CursorBench 3.2.0 | 73.4% | 70.5% | 70.0% | 67.2% |
| OSWorld 2.0 — partial | 77.9% | 72.9% | 75.4% | — |
| OSWorld 2.0 — strict | 41.7% | 36.1% | 39.6% | — |
| Humanity's Last Exam — no tools | 60.9% | 57.8% | 56.6% | — |
| Humanity's Last Exam — with tools | 65.0% | 63.8% | 63.6% | — |
| GDPval-AA v2 | 1,853 Elo | 1,723 | 1,824 | 1,711 |
The largest improvement is on Terminal-Bench-Science 0.1, where Fable 5.1 reaches 52.6%, compared with 24.7% for Fable 5. On GDPval-AA v2, Fable 5.1 reaches 1,853 Elo, ahead of Opus 5 at 1,824 and Fable 5 at 1,723.
Artificial Analysis independently reports that Fable 5.1 reaches a 66 Intelligence Index score at maximum effort, placing it ahead of Opus 5 at 63 and Fable 5 at 60 in that evaluation.
Benchmark results should nevertheless be interpreted by workload. For example, third-party benchmark aggregation reports Fable 5.1 at 89.1% on SWE-bench Multilingual versus 89.5% for Opus 5, illustrating that Fable 5.1 does not automatically dominate every specialized evaluation.
Claude Fable 5.1 vs Claude Fable 5 vs Claude Opus 5
| Model | Context | Max Output | Primary Position | Default Effort |
|---|---|---|---|---|
| Claude Fable 5.1 | 1M | 128K | Long-horizon reasoning and agentic work | High |
| Claude Fable 5 | 1M+ | 128K | Previous Fable generation | High |
| Claude Opus 5 | 1M | 128K | Complex agentic coding and enterprise work | High |
Fable 5.1 is the stronger choice when the workload involves long-running autonomous execution, demanding reasoning, multi-step research, or complex coding. Anthropic itself recommends starting most workloads with Opus 5 and moving to Fable 5.1 when higher-effort Opus 5 evaluations remain insufficient.
Compared with Fable 5, the major changes are improved agentic performance, lower cache-read costs, per-message effort controls, progress updates, and changes to thinking-block and tool-use behavior.
Limitations of Claude Fable 5.1
Fable 5.1 is not the fastest or cheapest Claude model. Anthropic classifies its comparative latency as slower, and its standard input/output rates are higher than Opus 5 and Sonnet 5.
It also has important API migration differences from Fable 5. In particular, forced tool use with tool_choice: any or tool returns a 400 error, while auto and none remain supported. Applications migrating from Fable 5 should also account for changes involving thinking blocks and conversation history.
Fable 5.1 also operates with production safety classifiers. Some restricted cybersecurity and biology requests can trigger refusal or fallback behavior, so benchmark results on tasks affected by safeguards should not be interpreted as unrestricted Mythos-class capability.
Best Use Cases for Claude Fable 5.1
- Repository-scale software engineering — multi-file refactoring, debugging, migration, testing, and long-running coding agents.
- Autonomous research — multi-step investigation requiring search, evidence collection, analysis, and synthesis.
- Large-document analysis — processing extensive technical documentation, contracts, reports, or research collections.
- Computer-use workflows — tasks involving desktop interfaces, spreadsheets, presentations, and other visual environments.
- Enterprise knowledge work — complex reports, financial or analytical workflows, and professional deliverables requiring sustained reasoning.
- Agentic systems — applications where the model must repeatedly call tools, inspect results, revise its plan, and continue until a task is complete.