Anthropic
Claude Fable 5
The smartest planner in the roster, but not the worker we trust to finish every build.
Canonical model record
Current identity, limits, and pricing
- Status
- Current
- API model ID
claude-fable-5- Context
- 1M
- Max output
- 128K
- API price / 1M tokens
- $10 input · $50 output
Visual prompt runs
Benchmark runs
Open each generated scene, or compare the same prompt across models.
Helm's Deep
Fortress siege scene testing scale, lighting, architecture, and cinematic atmosphere.
Hogwarts Broom Flight Simulator
Broom-flight scene testing depth, motion cues, castle scale, and fantasy mood.
Jabberwock
Dark fantasy encounter testing creature design, forest mood, and narrative staging.
Low Poly World
Stylized island build testing composition, color, and low-poly worldbuilding.
Office Life
Workplace vignette testing everyday scene logic, objects, and believable office detail.
Petri Dish
Microscopic ecosystem testing organic forms, scientific clarity, and cellular detail.
Universe Simulator
Cosmic system testing orbital structure, glowing bodies, scale, and simulation readability.
Vice City
Neon coastal city testing vehicles, architecture, atmosphere, and dense urban layout.
Yingzao Fashi Assembly
Timber assembly scene testing structure, joinery, construction order, and material clarity.
Official benchmark profile
How Claude Fable 5 scores beyond our visual tests.
Comparable numeric rows from the Anthropic system card and the shared GPT-5.6 comparison table. Marketing-only claims such as ViBench and spreadsheet speedups are excluded unless a score is published.
SWE-bench Pro
80.0%SWE-bench Verified
95.0%Terminal-Bench 2.1
84.3%Full official benchmark table14 rows with source settings and peer charts
| Benchmark | Area | Score | Setting / comparison |
|---|---|---|---|
| SWE-bench Pro | Coding | 80.0% | Average over 5 trials; standard configuration with thinking blocks included. |
| SWE-bench Verified | Coding | 95.0% | Average over 5 trials on the 500-problem verified subset. |
| Terminal-Bench 2.1 | Agentic coding | 84.3% | Anthropic mini-SWE-agent harness at high effort; OpenAI’s shared table reports Fable at 83.1%. |
| FrontierCode Diamond | Agentic coding | 29.3% score / 30.2% pass | Mean@5 at xhigh reasoning effort on Cognition’s Diamond subset. |
| FrontierCode Main | Agentic coding | 46.3% score / 48.8% pass | Mean@5 at xhigh reasoning effort on Cognition’s Main subset. |
| CursorBench | Agentic coding | 72.9% | Cursor production agent harness at maximum effort. |
| GPQA Diamond | Academic reasoning | 92.6% | Shared GPT-5.6 comparison table. |
| FrontierMath Tier 1-3 v2 | Math | 87.0% | Shared GPT-5.6 comparison table. |
| FrontierMath Tier 4 v2 | Math | 87.8% | Shared GPT-5.6 comparison table. |
| OSWorld-Verified | Computer use | 85.0% | Pass@1 averaged over 5 runs on 361 tasks with 100 action steps. |
| Blueprint-Bench 2 | Spatial reasoning | 38.6% | Andon Labs standard harness; normalized composite floor-plan score. |
| OfficeQA Pro | Document reasoning | 57.9% | Databricks image-based OfficeQA Pro evaluation. |
| Finance Agent Benchmark v2 | Finance | 56.31% | Vals AI evaluation with adaptive thinking and max effort. |
| MCP Atlas | Office work | 83.3% | Pass rate on real-world MCP tool-use workflows. |
Superbash commentary
Our take
Fable 5 remains the team’s S-tier choice for raw intelligence, conversation, planning, and difficult architecture. The important operating note is that it can be reluctant to execute and often routes coding work to Opus, so the team uses it as the brain of a workflow rather than assuming it will be the best hands-on builder.
Best for
Watch out
Why it is ranked here
Evidence and commentary
Superbash editorial model ranking
Takeaway: Claude Fable 5 is currently placed in Tier S.
The August 2026 editorial roster places Claude Fable 5 at rank 1.
Open source →NEW AI Model Tier List for Vibe Coding!
Takeaway: Fable 5 leads the S tier when access is available.
The source video ranks it as the strongest tested model for difficult coding work, while warning that access is unstable.
Open source →Keep learning