xAI·released 2025-07-091 source
Grok 4 benchmark scores: 19 benchmarks tracked. 26% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 0 of 19 benchmarks·best result 94.4 on Fiction.LiveBench (unverified)·5 of 19 independently reproduced·$3/$15 per M tokens
Consensus: LiteLLM · See every model’s pricing →
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| Fiction.LiveBench | 94.4accuracy (%) | unverifiedT1 | 2025-07-09Fiction.live leaderboard |
| OTIS Mock AIME 2024-2025 | 83.98accuracy (%) | reproduced· optimizedT1 | 2025-07-09Epoch AI |
| GPQA diamond | 82.67accuracy (%) | reproduced· optimizedT1 | 2025-07-09Epoch AI |
| Aider polyglot | 79.6accuracy (%) | unverified· optimizedT1 | 2025-07-09Aider LLM Leaderboards |
| Lech Mazur Writing | 76.9accuracy (%) | unverifiedT1 | 2025-07-09lechmazur/writing Github repository |
| ARC-AGI | 66.67accuracy (%) | unverified· optimizedT2 | 2025-07-09 |
| METR Time Horizons | 66.58accuracy (%) | unverifiedT1 | 2025-07-09METR - Measuring AI Ability to Complete Long Tasks |
| SimpleBench | 52.6accuracy (%) | unverifiedT1 | 2025-07-09https://lmcouncil.ai/benchmarks |
| DeepResearch Bench | 47.3accuracy (%) | unverifiedT1 | 2025-07-09https://evals.futuresearch.ai/#drb |
| WeirdML | 45.73accuracy (%) | unverifiedT1 | 2025-07-09https://htihle.github.io/weirdml.html |
| GeoBench | 45accuracy (%) | unverifiedT1 | 2025-07-09GeoBench leaderboard |
| Balrog | 43.6accuracy (%) | unverifiedT1 | 2025-07-09https://balrogai.com/ |
| Cybench | 43accuracy (%) | self-reported· optimizedT1 | 2025-07-09Grok 4 Model Card |
| FrontierMath-2025-02-28-Private | 34.48accuracy (%) | reproduced· optimizedT1 | 2025-07-09Epoch AI |
| Terminal Bench | 27.2accuracy (%) | unverified· optimizedT1 | 2025-07-09Terminal-Bench v2 Leaderboard |
| Chess Puzzles | 24.24accuracy (%) | reproducedT1 | 2025-07-09Epoch AI |
| GDPval | 21.1accuracy (%) | unverifiedT2 | 2025-07-09 |
| ARC-AGI-2 | 15.97accuracy (%) | unverifiedT2 | 2025-07-09 |
| FrontierMath-Tier-4-2025-07-01-Private | 3.47accuracy (%) | reproducedT1 | 2025-07-09Epoch AI |
Claims drawn from cited facts, not live model generation.
Developed by xAI, this model dates to 2025.
2 cited facts
This model is scored on 19 tracked benchmarks, but it currently holds no top score on any of them. Its record can be trusted: the scores have been independently reproduced, meaning outside evaluation backs the numbers up rather than leaving them as unconfirmed vendor claims.
3 cited facts
Its mean gap to the best recorded score across its tracked benchmarks is 29.87 points.
1 cited fact
5 facts cross-checked across data sources: 5 single-source.
Source facts, citations, and refresh stamp for this record.
Sources: epoch_benchmarks, litellm_prices
Grok 4 is an AI model developed by xAI, released 2025-07-09. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
Grok 4 has recorded scores on 19 benchmarks, each shown with its evidence status.
Grok 4 has recorded scores on 19 benchmarks — Lech Mazur Writing, GPQA diamond, OTIS Mock AIME 2024-2025, WeirdML, Chess Puzzles, Cybench, and 13 more. The full table above shows each score with its evidence status.
5 of 19 recorded scores (26%) are independently reproduced rather than self-reported by the lab.
Grok 4 has 19 tracked claims: 5 independently reproduced, 1 self-reported, 13 unverified.
Listed API pricing: $3 per million input tokens, $15 per million output tokens. See the pricing block for the full breakdown.