Claude Fable 5.2
Claude model for demanding reasoning and long-horizon agentic work
Shogo Agent Index
Supported#1 of 9276.4range 71.7–81.2 · 32% of weight measured
Agentic coding
86
26% measured
Long-horizon autonomy
78
50% measured
Knowledge-work agents
67
52% measured
Computer use
77
estimated
Security & context learning
65
estimated
Agentic benchmarks
| Benchmark | Score | Normalised | Index weight | Harness / source |
|---|---|---|---|---|
| SWE-bench Verified | est. | 8% | predicted 92 | |
| Terminal-Bench 2.0 | est. | 8% | predicted 98 | |
| DeepSWE | est. | 6% | predicted 88 | |
| FrontierSWE | 69.9% | 5% | proximus · $168.78/task | |
| CursorBench | 60.0% | 4% | source · $32.79/task | |
| GSO-Bench | est. | 4% | predicted 88 | |
| METR Time Horizons | est. | 10% | predicted 90 | |
| Vending-Bench 2 | $9,628 | 6% | source | |
| PostTrainBench | 44.6% | 4% | Claude Code | |
| APEX-Agents | 80.1% | 8% | source | |
| GDPval | est. | 7% | predicted 78 | |
| Remote Labor Index | 41.3% | 5% | source | |
| DeepResearch Bench | est. | 5% | predicted 55 | |
| OSWorld 2.0 | est. | 6% | predicted 63 | |
| OSWorld | est. | 4% | predicted 98 | |
| Cybench | est. | 5% | predicted 97 | |
| CL-bench | est. | 5% | predicted 34 |
Other benchmarks
| Benchmark | Category | Score | Source |
|---|---|---|---|
| ARC-AGI | Reasoning | 98.5% | source |
| ARC-AGI-2 | Reasoning | 93.8% | source |
| Chess Puzzles | Reasoning | 79.0% | source |
| CritPt | Reasoning | 35.7% | source |
| DTBench | Knowledge | 99.2% | source |
| FrontierMath Tiers 1-3 | Math | 95.1% | source |
| FrontierMath Tier 4 | Math | 100.0% | source |
| GPQA Diamond | Reasoning | 96.0% | source |
| LMCA | Knowledge | 70.6% | source |
| OTIS Mock AIME | Math | 99.7% | source |
| ProofBench | Math | 100.0% | source |
| SciCode | Coding | 67.7% | source |
| SimpleBench | Reasoning | 94.9% | source |
| SimpleQA Verified | Knowledge | 94.0% | source |
| WeirdML v3 | Coding | 43.1% | claude_code |