GPT-6.2 Astra lands #2 on code-review benchmark MacroscopeBench at 1/5 the cost of Opus 5.5
film_girl · x · 2026-11-20
MacroscopeBench, which tests whether models catch real bugs at the commits that introduced them, has added OpenAI's GPT-6.2 Sol and GPT-6.2 Astra.
Key results:
- GPT-6.2 Astra (high effort) scored 83.2, second only to Claude Opus 5.5 at max effort (83.9) and #1 among high-effort runs, with 90% precision but 77.4% recall — at $2.17 per review in 3 minutes.
- GPT-6.2 Sol (high effort) ranked #4 overall (81.7) and is the most precise model tested at 91.6%, costing just $0.69 in 2m19s.
- The top-ranked Opus 5.5 max effort run costs $11.58 and takes over 10 minutes per review.
Takeaway: on precision-heavy code review, the GPT-6.2 family nearly matches Opus 5.5's combined score at a fraction of the cost and latency.
More from coding & agent
- Rex now works over SSH with one-command server install — josh_wills · 2026-11-20
- Seroter Daily #913: Always-on Agents, Envoy Agentic Stack, Tool Retrieval in Gemini API — rseroter · 2026-11-20
- Blogger's Take: GPT-6.2 Astra as Orchestrator Plus Gemini 4 Neon Is the Best Coding Combo — haider1 · 2026-11-20
- Software factories underdeliver: revisiting the 'autonomous product org' loop design vision — pzakin · 2026-11-20
- Firecrawl goes local: Rust rewrite scrapes 37x faster with 11MB idle memory, fully open source — devdigest · 2026-11-20
- Stripe's Fraud Model Gets 39% Higher Recall by Training on Explanations in the Right Order — emilygsands · 2026-11-20