Kimi K3 report adds in-house coding, agent, and WebDev benchmark tables
nrehiew_ · x · 2026-07-29
This follow-up adds two tables from the Kimi K3 report.
In-house benchmark table
It compares Kimi K3 (max) with several proprietary and open-weight models across coding, agent, and conversational benchmarks. Notable entries include:
- Coding Experience: Kimi K3 scores competitively across Claude Code and Codex harnesses.
- General Agent Experience: strong results on benchmarks such as 24/7 ClawBench 2.0, MIRA Bench, KAET, CLIF Bench, Agentic Vision Bench, Swarm Bench, Online Experience, Deep Research Bench, Finance Bench, KWV Bench, DECK Bench, and Agent Behavior Bench.
- Conversational Experience: Kimi K3 performs strongly on Faithfulness and Chat All-in-One Bench.
WebDev bench
A second table compares Kimi K3 (max) against Claude Opus 4.8 (max) under blind expert judging on:
- Games
- 3D / WebGL / Shader
- Website / UI Clone
- Overall
The report says experts judged outputs without knowing which model produced them, scoring code quality, feature completeness, visual fidelity, and interaction experience.
More from Infra
- SkyRL ships Day-0 RL support for Beam with +62% LoRA decode throughput in vLLM — vllm_project · 2026-10-16
- DFlash2 delivers up to 3.03x decode-only speedup for Beam across five benchmarks — vllm_project · 2026-10-16
- vLLM's Beam serving kernels cut GB200 forward time from 8.15 to 6.47 ms at TP8 batch 128 — vllm_project · 2026-10-16
- Reflection's 501B MoE Beam hits 463 tok/s on 8x GB200 with NVFP4 at day 0 — zhyncs42 · 2026-10-16
- Nemotron 3.5 Bolt runs 16 concurrent streams on a single DGX Spark — NVIDIAAI · 2026-10-16
- Why your computer can't store 0.2: an IEEE 754 primer — Franc0Fernand0 · 2026-10-16