Kimi K3.1 Pro gains less from test-time compute than Opus 5.0, data shows
teortaxesTex · x · 2026-11-05
Data shared by pzeroresearch: Kimi K3.1 Pro is strong at low compute budgets, scoring about 1.06x at 15 H100 hours, close to Opus 5.0. But at 40 H100 hours its multiplier falls to 0.78x while Opus 5.0 rises to 1.3x. Takeaway, as teortaxesTex notes: new frontier models are much better at making use of test-time compute, and K3.1 Pro extracts far less value from extra compute than the newest Opus models.
Related event: Kimi K3.1 Pro Scores 0.78x Compute Multiplier in TasteVal Benchmark(2 posts)→
More from Models
- Claude trains a model instead of writing code: 1257-param pixel-art generator in 10 minutes — victormustar · 2026-11-05
- MoE beats dense in matched OOD finetune: DeepSeek V3.1 hits 0.804 CE vs Llama 405B — kalomaze · 2026-11-05
- Grok Bot tests sharing disclosure: files, skills and credentials visible before you share — XFreeze · 2026-11-05
- Ari agent post-trains Qwen3.5-9B unsupervised for 16h, reward 0.118→0.629 — ypatil125 · 2026-11-05
- Happenstance scores 84.7 on PeopleSearchBench's 119 queries, nearly 20 points above next best — alex_teichman · 2026-11-05
- Google Research unveils Agile Deliberation, an LLM framework for teaching VLMs subjective categories — RanjayKrishna · 2026-11-05