Arena weekly: Kimi K3.1 Pro tops open models, Gemini leads video, GPT Image 2x faster
arena · x · 2026-11-07
LMArena's weekly roundup:
- Kimi K3.1 Pro (Moonshot): open weights released, now #1 among open models in Agent Arena with a +5.3% net improvement.
- Hy4 (Tencent Hunyuan): debut as the top Chinese model in Agent Arena, +5.4% net improvement at a $0.57 median cost per task, reshaping the Pareto frontier; open weights coming soon.
- Inkling 2 Small (thinkymachines): scored 1648 pts in Code Arena: WebDev; would be #1 among US open models once weights drop.
- Gemini Omni 1.2 Flash (Google DeepMind): debuted #1 in both Text-to-Video (1541) and Image-to-Video (1502).
- GPT Image 2.6 Sunburst/Flare (OpenAI): nearly 2x the speed of GPT Image 2.5, with Sunburst (Medium) at #2 in Text-to-Image (1409 pts).
More from Models
- XBow Benchmarks AI Models by Cost: Claude Opus 5.5 Leads at 91.5% Solved Under $15 — mboehme_ · 2026-11-07
- Thinking Machines' Inkling-2-small tops US open-weight models on weight-space coding — keirp1 · 2026-11-07
- Vitalik Buterin: 'Open weight is not open source' on AI models — wonderwomancode · 2026-11-07
- Zellige AI open-sources Tessera-27B-B-CJK-RL: RL-boosted CJK-thinking coder cuts thinking tokens by a third, matches base on hard math — ghgi_ · 2026-11-07
- moyix: I'm almost scared to find out how good LLMs are at record/replay debugging now — moyix · 2026-11-07
- Steering Qwen's layer-15 attention output makes it claim to be a group of people — Sauers_ · 2026-11-07