Community SEAL-based AI leaderboard: GPT-6-Astra tops at 75.58 after coverage adjustment
Hrstar1 · reddit · 2026-10-31
A Reddit user compiled an overall ranking from Scale Labs' SEAL test results, applying an adjusted score to penalize models tested in few categories.
Top of the board:
- GPT-6-Astra: 75.58 (12/15 coverage)
- Claude-Fable-5.1: 74.30 (12/15)
- Gemini-4-Argon: 72.06 (11/15)
- Claude-Fable-5: 58.41; 5. Claude-Opus-5: 57.62
Other notes:
- Clear gap between the top three (72+) and the rest (below 59)
- Best Chinese/open model is DeepSeek V4.1 Flash 1017 at #15; GLM-5.2/5.3 at #20-21, Kimi-K3 at #24, Qwen3.8 at #29; Mistral Large 4 only #58
- Covers 119+ models including GPT-OSS-120B (#41) and MiniMax-M3 (#79)
Caveat: this is a community-made weighted aggregation, not an official leaderboard, and coverage differences heavily affect scores.
More from Models
- Inference optimization race heats up: practitioners now want fast, cheap 'good enough' models, not the absolute best — bendee983 · 2026-10-31
- OpenAI tightens Codex account limits: stacking paid accounts on one device can now get you locked out — TheZachMueller · 2026-10-31
- NVIDIA Releases Nemotron 3.5 Bolt E4B: 4GB On-Device Model with 80% Fewer Output Tokens — heatOverflower · 2026-10-31
- Epoch index shows open-weight models closing the gap with closed frontier to under 10 points — SpencrGreenberg · 2026-10-31
- Gemini 4 Carbon hits new SOTA on Riemann-Bench with 0.8 Pass@1, beating GPT-5.6 Sol — NielsRogge · 2026-10-31
- Anthropic's fable 5.5 reportedly delayed, possibly overshadowed by Opus 5.5 — haider1 · 2026-10-31