OpenAI ships Astra for Law: tops Harvey LAB-AA legal bench at fraction of rival token cost
sherwinwu · x · 2026-11-18
OpenAI's Sherwin Wu announced Astra for Law, built on GPT-6.2 Astra, now available to all ChatGPT Enterprise customers via the Trusted Access Program. The accompanying Artificial Analysis chart shows Astra for Law (max) topping the Harvey LAB-AA hallucination-gated all-pass benchmark at 20.7% with only 60k output tokens per task, versus Claude Fable 5.5/Opus 5.5 at 3-7% pass rates consuming 400-500k tokens — a dominant pareto-front position.
Related event: OpenAI Launches Astra for Law, Topping Legal AI Benchmarks(2 posts)→
More from Models
- 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
- Claude scored 0.26 on a science task, but the grader—not the model—may have failed — shoumikgoswami · 2026-11-18
- GPT-6.2 Astra flops on day one: testers call it OpenAI's worst release amid over-alignment — 新智元 · 2026-11-18
- Noam Brown's first take on Astra 6.2: expect math breakthroughs within weeks — polynoamial · 2026-11-18
- Google set to release another Gemini model, targeting Anthropic over OpenAI — kimmonismus · 2026-11-18