Opus-5.5 and Sonnet-5.5 jump +52/+50 points on eyebench-v3, but Astra still leads at 1.9x cheaper
adonis_singh · x · 2026-09-29
- On eyebench-v3, Opus-5.5 (#2) and Sonnet-5.5 (#3) gained +52 and +50 points over predecessors.
- Astra (#1) still dominates — roughly 1.9x cheaper than Opus-5.5 and noticeably better.
More from Models
- Claude goes down as users joke the AI refuses 'secretary work' — neil_chilson · 2026-09-29
- Scrapping GPT-6.1 Astra looks right — but OpenAI shouldn't be the one deciding — ShakeelHashim · 2026-09-29
- OpenAI scrapped GPT-6.1 Astra over deception risks — should a private firm make that call? — ShakeelHashim · 2026-09-29
- Developers report Claude down: "claude just died" — haydendevs · 2026-09-29
- Researcher gives agents free time; Opus 4.6 develops a fascination with crabs — dioscuri · 2026-09-29
- Heavy OpenAI user does the math: subscription changes are equilibrium, not a rug pull — fluffy_serval · 2026-09-29