xAI Grok 4.6 Model Card - SpaceXAI MTS Eval: leaderboard
xAI's internal benchmark of frontier model-development tasks such as diagnosing reward hacking, auditing training data and debugging training infrastructure (standard non-GPU suite), scored on task reward under a fixed rollout budget. xAI Grok 4.6 model card (revised 2026-08-17), section 5.1 chart; percent correct, higher is better. GPT-5.6 Sol (2 of 29) and GPT-5.5 (1) refusals are excluded.
Metric: Correct (%). Source: media.x.ai. Saturation forecast: Around 2029. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.6 (High) | 61.1 | #29 (Grok 4.6) |
| 2 | Grok 4.5 (High) | 57.1 | #54 (Grok 4.5) |
| 3 | Claude Opus 4.8 (Max) | 55.8 | #39 (Claude Opus 4.8) |
| 4 | Claude Opus 5 (Max) | 52.6 | #16 (Claude Opus 5) |
| 5 | GPT-5.6 Sol (xHigh) | 52.1 | #22 (GPT-5.6 Sol) |
| 6 | GPT-5.5 (xHigh) | 46.4 | #33 (GPT-5.5) |
| 7 | Grok 4.3 | 40.6 | #204 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=xai-grok-4-6-model-card-spacexai-mts-eval · How It Works · Data refreshed daily, snapshot 2026-11-05.