Sol 6.1 hits perfect benchmark score with half the tokens of Sol 6, tester finds
startassets · reddit · 2026-09-30
A community tester benched Sol 6.1 and concludes it 'might be what Sol 6 was supposed to be.'
- Context: Sol 6 shipped claiming 50% API cost reduction but real-world performance dropped significantly, confirmed by multiple benchmarks; a $200 plan quota cut made things worse
- Results: Sol 6.1 hit a 100 benchmark score with speed and precision, using half the tokens of Sol 5.6 and Sol 6
- Caveats: behavior on real tasks and quota drain rates remain untested; Pi didn't support the model yet, so it was added manually—full cost data to come in reruns
It's a community-driven benchmark; repo on GitHub and dataset viewer on Hugging Face Spaces, with contributions welcome.
More from Models
- Will user protest make Anthropic revert the Pro 200x usage cut? — Kakachia777 · 2026-09-30
- Claude Chat still on Sol 5.6, and Sol 6.1 isn't coming either — throwawaysusi · 2026-09-30
- Grok Bot now lets you customize your profile picture with uploads or a prompt — XFreeze · 2026-09-30
- Local 2-bit quantized model catches a logic trap that frontier DeepSeek V4-Flash fell for — HolidayBit143 · 2026-09-30
- Claude's Free Limit Reset Button Appears Only After Subscribing to Max, Not Pro — ThePeterMick · 2026-09-30
- Procedural cheetah in Three.js: GPT 6.1 Sol Max vs Opus 5.5 Max — majidmanzarpour · 2026-09-30