Arena weekly: Claude Fable 5.2 tops Agent Arena, Gemini 4 Hydrogen fastest
arena · x · 2026-11-14
LMArena's weekly roundup:
- Claude Fable 5.2 (Max) (Anthropic) takes #1 in Agent Arena with a +30.5% net improvement across 7.8k real-world sessions.
- Gemini 4 Hydrogen (High) (Google DeepMind) debuts as the fastest top-10 Agent Arena model at 7.3s per typical step.
- Muse Ember (Max) (Meta's smallest model) enters Code Arena: WebDev at #37 (1609 points, $0.03 median cost per task).
- Inkling 2 Small (Thinking Machines) is now open weights and the #1 US open model in Code Arena: WebDev (1647).
- MiniMax H3-Image is the #1 open model in Multi-Image Edit (1365, #11 overall), on par with Nano Banana 2/Pro.
- ERNIE 5.5 Flash Preview (Baidu) trends toward #2 open model in Code Arena: WebDev (1668, AutoEval).
More from Models
- Chart: Google models always say they 'spawned in,' everyone else 'woke' — liminal_bardo · 2026-11-14
- Microsoft's MAI-Transcribe-2 nears the top on speed: 93.7 seconds of audio per second — ArtificialAnlys · 2026-11-14
- Artificial Analysis launches speaker diarization benchmark; NVIDIA Nemotron 3.5 leads at 3.3% WDER — ArtificialAnlys · 2026-11-14
- Tandem-vLLM 0.2.0 runs DeepSeek-V4.1-Flash on 3090/4090-class GPUs with 2-bit expert offload — smflx · 2026-11-14
- AssemblyAI Adds GPT-6.1 Terra, Beam-1, DeepSeek V4.1 Flash to LLM Gateway — AssemblyAI · 2026-11-14
- Continual Learning Won't Block Full Research Automation — but Will Block Enterprise Adoption — ChrisGPT · 2026-11-14