Personal Benchmark: Opus 5.5 Hits Fable Level, GPT-5.6-Sol Still Leads
danshipper · x · 2026-09-23
Dan Shipper of Every points to @hammermt's personal benchmark Mike's Checks (9 models, 15 real work tasks) for a real-world read on Opus 5.5.
Highlights: Opus 5.5 scores 80% overall, matching Claude Fable 5 (79%) on many tasks, with 100% on pptx and roleplay-interview, though it timed out on a few tasks. GPT-5.6-Sol leads at 84%, Grok-4.6 hits 83%, and DeepSeek-V4-Flash trails at 58%.
The benchmark answers "which model is best at my job," scored via scripted checks plus an LLM judge, not a general capability claim.
Related event: Anthropic Launches Claude Opus 5.5: Stronger, 40% Cheaper, Tops AA Index(259 posts)→
More from Models
- Claude's connector management criticized: Chat-only visibility, buried in 'Customize' tab — tristanbob · 2026-11-09
- Devs: Sora makes indistinguishable video while image gen still looks like slop — Angaisb_ · 2026-11-09
- Rumor: NDA on rumored fable 5.5 model reportedly ending, launch may land next week — iruletheworldmo · 2026-11-09
- GameFeel Bench v1.4: Qwen-4-Flash-Early leads at 49.9 after score fix — Sea_Emergency_8458 · 2026-11-09
- Google AI Studio shares new tutorial and more info — DuRuofei · 2026-11-09
- Indian "built in India" Alpie 9B exposed as a near bit-for-bit copy of Qwen3.5-9B — dotey · 2026-11-09