Up to four models
Claude Fable 5 vs MiniMax-M3
On the index (GPT-4 = 100), Claude Fable 5 scores 183 and MiniMax-M3 147. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Claude Fable 5
183167–199
WeightsClosed
ReleasedJun 2026
Price in / out$10 / $50
Context1M
Core tests taken6 of 9
147131–164
WeightsOpen
ReleasedJun 2026
Price in / out$0.30 / $1.2
Context1.05M
Core tests taken4 of 9
Claude Fable 5 leads MiniMax-M3 by 36 points, ahead on 4 of 4 shared core tests, at 38× the price.
Test by testthird-party results
Claude Fable 5
MiniMax-M3
Every result as a table
| Test | Claude Fable 5 | MiniMax-M3 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 93.9% | 90.9% |
| LiveBench Data Analysis | 80.5% | 76.2% |
| LiveBench IF | 75.8% | 57.5% |
| LiveBench Language | 90.7% | 76.8% |
| LiveBench Reasoning | 89.6% | 74.5% |
| GPQA Diamond (Vals AI) | 93.2% | 92.7% |
| MMLU Pro | 91.5% | 84.2% |
| MMMU Pro | 89.3% | 81.2% |
| SimpleBench | 81.9% | 45.8% |
| Maths | ||
| AIME-style maths (OTIS mock) | 100.0% | 71.1% |
| LiveBench Mathematics | 96.0% | 77.0% |
| ProofBench v1.1 | 95.0% | 18.0% |
| Chess puzzles | 47.0% | 17.0% |
| Surface Evolver Bench | 95.0% | 55.0% |
| Coding | ||
| Terminal-Bench 4.0 | 41.4% | 1.0% |
| LiveBench Agentic Coding | 62.2% | 40.7% |
| LiveBench Coding | 86.0% | 68.2% |
| Code Migration | 55.1% | 19.9% |
| LiveCodeBench | 89.8% | 82.2% |
| SWE-bench | 95.0% | 75.0% |
| Vibe Code Bench v1.1 | 90.3% | 47.6% |
| Agents | ||
| APEX-Agents | 63.6% | 37.7% |
| DTBench | 98.4% | 78.9% |
| LMCA | 61.2% | 33.7% |
| Mystery game puzzles | 52.0% | 8.0% |
| Professional work | ||
| Finance Agent (v2) | 56.3% | 48.3% |
| CorpFin v2 | 71.8% | 68.1% |
| EMB | 73.7% | 47.8% |
| LegalBench | 88.6% | 85.4% |
| Legal Research Bench | 49.5% | 29.8% |
| MedCode | 56.1% | 46.3% |
| MedScribe | 88.5% | 87.3% |
| MortgageTax | 68.9% | 68.4% |
| Public Benefits Bench v1.1 | 70.4% | 64.1% |
| Tax Agent Bench | 38.3% | 20.6% |
| TaxEval v2 | 76.9% | 72.7% |
Tests at least two of these models have taken. Bold rows are the core tests.