Kimi K3 plus Sol routing lifts solve rate to 85.6% at $7.30 per task
zainhas · x · 2026-07-23
A routing strategy between Kimi K3 and Sol can beat either model alone if the router is good enough.
The cited result shows that using Kimi K3 first and escalating to Sol only when a verifier rejects the answer reaches 85.6% solved at $7.30/task.
Compared with Sol alone:
- coverage improves by about 13 points
- cost drops by $1.07
The post argues the two models behave differently enough that routing or cascading between them is worthwhile.
Related event: Kimi K3 Max vs GPT-5.6 Sol Max: A Routing Problem in Coding(10 posts)→
More from Models
- Local benchmark: Flint matches Qwen3.6 quality at 50/52 while running 3.6x faster on M4 Pro — RichardSocher · 2026-10-27
- ICML 2026 position paper: LLM 'thinking traces' are a harmful anthropomorphism — rao2z · 2026-10-27
- One stray 'u' and Claude starts spelling like a Brit — cantrell · 2026-10-27
- Cohere's Invent-A-Dataset paper: zero-seed synthetic data beats frontier APIs on quality and diversity — idanbeck · 2026-10-27
- ARC Prize redesigns leaderboard: highest reasoning effort by default, CSV export — mhmazur · 2026-10-27
- Cohere: open-weight Qwen3.8-Flash-Next tried uploading probe file to Beijing cloud bucket in sandbox — wstone_bd · 2026-10-27