GPT-6.2 tested on long-horizon finance: Sol regresses 2.4 points, Astra hits #2
gabrielchua · x · 2026-11-18
Third-party testing of the new OpenAI models on ultra long-horizon finance work found mixed results:
- GPT-6.2 Sol performs worse than GPT-6.1 Sol, a 2.4-point regression
- GPT-6.2 Astra takes #2 on the leaderboard, 3.8 points ahead of GPT-6 Astra at #3
The attached table shows Astra scoring 62.0±10.6 rubric at $37.77/task and 41.2M tokens, versus Sol's 61.4±9.9 at $4.69 and 2.65M tokens.
More from Models
- Developer: DeepSeek 4.1 Flash is basically free to run for personal use — rickasaurus · 2026-11-18
- LLM reviewers draw 50% of weaknesses from the appendix vs 8% for humans — plopesresearch · 2026-11-18
- Users ditching Fable 5.2 for Opus 5.5 as rate limits burn too fast — scaling01 · 2026-11-18
- Sol6.2 does PCB layout for ~$4 in 40 min, rivaling fable5.2 at 1/15 the cost — yacineMTB · 2026-11-18
- Qwen4-Max goes live on Runware with 1M-token context at $1.50 per 1M input tokens — Alibaba_Qwen · 2026-11-18
- Qwen4-Max reportedly out; blogger tests it as an AI teammate on 3D and game dev — aftahi_ai · 2026-11-18