OpenAI's GPT-6.2 Astra Tops Harvey Legal Eval at a Third of the Cost
ahelkky · x · 2026-11-19
OpenAI released GPT-6.2 Astra, a frontier model pitched for the hardest projects, setting new highs on evals across professional work and scientific research at less than a third of the per-task cost of GPT-6 Astra.
OpenAI researcher @zeyuanxy said the team was not near the top on legal evals a few months ago, but Astra is now best on the Harvey LAB-AA legal benchmark at a fraction of the cost. OpenAI says the gains show up in slideshows, websites and 3D designs.
More from Models
- Astra 6.2 hands-on: rambling about an idea yields a working app — kagigz · 2026-11-19
- Every's Vibe Check: GPT-6.2 Astra and Sol handle bounded tasks but fall short on judgment — danshipper · 2026-11-19
- Microsoft releases Aion-1.0-Instruct-2.3B, a small on-device model — adefa · 2026-11-19
- Cascade MoE on Bittensor beats Salesforce's Moirai2 on GIFT-Eval 19 days after launch — const_reborn · 2026-11-19
- Sciforium hits 1,129 output tokens/sec P90 on DeepSeek-V4.1-Flash over 4 AMD MI355X GPUs — rohanpaul_ai · 2026-11-19
- Beyond beating games: Canto Reality unveils a model learning to understand play — xennygrimmato_ · 2026-11-19