Kimi K3.1 Costs ~3x DeepSeek Flash Per Task but Nearly Doubles Terminal-Bench 4 Score
bookwormengr · x · 2026-10-24
- Terminal-Bench 4 cost-vs-score chart: Kimi K3.1 Pro (max) scores 65% at $4.5/task, vs DeepSeek-V4.1-Flash-1017 at 34% and $1.7/task — roughly 8x the price for nearly double the score.
- The chart also plots Claude Opus 5.5, Claude Sonnet 5.5 ($22/task), GPT-6 Astra max, and GLM-5.3.
- The author argues Kimi is worth it for coding/content use cases, speculates Moonshot has a private frontier-cyber-capable model, and hopes DeepSeek and GLM follow Moonshot's responsible example by not releasing such models unsafeguarded into the wild.
More from Models
- Moonshot removed cyber-related training data from open-weight Kimi K3.1 Pro — nrehiew_ · 2026-10-24
- User says Opus silently downgraded to 4.8 mid-task, then told him to talk to people — natesiggard · 2026-10-24
- Physical-engineering benchmark: Claude Opus 5.5 tops at 48.3%, Grok 4.7 trails at 30.1% — teortaxesTex · 2026-10-24
- Suspicious new GitHub accounts reference Qwen4 in Alibaba T-Head repo PRs — brainExploded99 · 2026-10-24
- Kimi K3.1 Pro takes on the pelican-on-a-bicycle 3D test: beats Opus 5, trails 5.5 — blaind0 · 2026-10-24
- Kimi's $19 coding plan roasted: 5-hour limit hit before a single meaningful line of code — teortaxesTex · 2026-10-24