Up to four models
GPT-6 Astra vs Mistral Large 4
On the index (GPT-4 = 100), GPT-6 Astra scores 199 and Mistral Large 4 160. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
GPT-6 Astra
199182–215
WeightsClosed
ReleasedSep 2026
Price in / out$10 / $50
Context1.05M
Core tests taken7 of 9
Mistral Large 4160144–177
WeightsOpen
ReleasedOct 2026
Price in / out$1.36 / $4.18
Context1.05M
Core tests taken4 of 9
GPT-6 Astra leads Mistral Large 4 by 38 points, ahead on 3 of 4 shared core tests, at 9.7× the price.
Test by testthird-party results
GPT-6 AstraMistral Large 4
Mistral Large 4Every result as a table
| Test | GPT-6 Astra | Mistral Large 4 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 95.8% | 88.1% |
| SimpleQA Verified | 75.6% | 20.0% |
| Maths | ||
| FrontierMath (tiers 1-3) | 93.7% | 58.6% |
| AIME-style maths (OTIS mock) | 100.0% | 98.3% |
| ProofBench v1.1 | 99.0% | 10.0% |
| Chess puzzles | 74.0% | 33.0% |
| FrontierMath (tier 4) | 97.6% | 26.8% |
| Coding | ||
| Terminal-Bench 4.0 | 59.6% | 27.8% |
| Code Migration | 67.7% | 30.6% |
| IOI | 100.0% | 45.3% |
| Vibe Code Bench 1-100 | 27.6% | 14.3% |
| Vibe Code Bench v1.1 | 89.6% | 78.4% |
| Agents | ||
| BioMysteryBench | 79.3% | 67.0% |
| DTBench | 97.3% | 88.2% |
| Furniture assembly | 80.0% | 20.8% |
| LMCA | 64.4% | 35.8% |
| Mystery game puzzles | 84.0% | 17.0% |
| Professional work | ||
| Finance Agent (v2) | 53.5% | 54.7% |
| EMB | 71.7% | 56.2% |
| HLAB-Vals-1.1 | 7.1% | 17.5% |
| Legal Research Bench | 39.4% | 31.7% |
| MedCode | 48.5% | 40.7% |
| MedScribe | 87.9% | 80.4% |
| Tax Agent Bench | 20.7% | 30.4% |
| Novel problems | ||
| MysteryMechanism | 53.1% | 12.2% |
Tests at least two of these models have taken. Bold rows are the core tests.
