Up to four models
DeepSeek-V4.1-Flash vs Grok 4.7
On the index (GPT-4 = 100), DeepSeek-V4.1-Flash scores 170 and Grok 4.7 168. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
DeepSeek-V4.1-Flash170153–186
WeightsOpen
ReleasedSep 2026
Price in / out$0.14 / $0.42
Context1.05M
Core tests taken6 of 9
Grok 4.7
168151–184
WeightsClosed
ReleasedSep 2026
Price in / out$2.0 / $6.0
Context500K
Core tests taken6 of 9
DeepSeek-V4.1-Flash and Grok 4.7 are too close to call (2 points apart, within the uncertainty), ahead on 3 of 6 shared core tests, and costs 14× less.
Test by testthird-party results
DeepSeek-V4.1-FlashGrok 4.7Every result as a table
| Test | DeepSeek-V4.1-Flash | Grok 4.7 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 89.8% | 92.7% |
| SimpleQA Verified | 42.9% | 56.0% |
| Maths | ||
| FrontierMath (tiers 1-3) | 67.4% | 53.0% |
| AIME-style maths (OTIS mock) | 98.3% | 98.1% |
| ProofBench v1.1 | 54.0% | 26.0% |
| Chess puzzles | 48.0% | 41.0% |
| FrontierMath (tier 4) | 26.8% | 17.1% |
| Coding | ||
| Terminal-Bench 4.0 | 19.7% | 28.8% |
| Code Migration | 45.6% | 44.8% |
| CyberBench v1.1 | 73.7% | 69.5% |
| IOI | 40.3% | 57.7% |
| ProgramBench | 0.5% | 0.5% |
| Vibe Code Bench v1.1 | 84.7% | 86.2% |
| Agents | ||
| APEX-Agents | 39.5% | 54.6% |
| BioMysteryBench | 67.8% | 69.3% |
| SRE Bench | 0.8% | 3.8% |
| Terminal-Bench-Science | 0.0% | 10.0% |
| DTBench | 89.9% | 96.0% |
| Furniture assembly | 34.2% | 20.8% |
| LMCA | 47.0% | 49.4% |
| Mystery game puzzles | 43.0% | 29.0% |
| Professional work | ||
| Finance Agent (v2) | 53.5% | 52.3% |
| EMB | 57.2% | 67.0% |
| LegalBench | 83.3% | 84.4% |
| Legal Research Bench | 41.3% | 47.1% |
| MedCode | 41.2% | 49.5% |
| MedScribe | 85.5% | 89.4% |
| Public Benefits Bench v1.1 | 64.3% | 65.6% |
| SAGE | 47.9% | 40.8% |
| Tax Agent Bench | 26.6% | 34.2% |
| Novel problems | ||
| ARC-AGI-2 | 72.9% | 61.4% |
| ARC-AGI-1 | 94.5% | 90.2% |
| ARC-AGI-3 | 0.9% | 1.8% |
| MysteryMechanism | 21.2% | 25.2% |
Tests at least two of these models have taken. Bold rows are the core tests.
