Observing Factuality Trends in Open-Source Models
arena · x · 2026-07-16
Arena continues to share observations from its factuality research.
- After increasing the factuality weight, most open-source models see a drop in their scores, particularly nvidia-nemotron-3-ultra.
- A few exceptions include mistral-medium-3.5 and Tencent huyuan-hy3-preview, which actually improve under higher factuality weights.
- They also observed that as flagship models continue to be released over time, OpenAI is the only vendor consistently improving pure factuality scores over a longer period; recently released models from xAI and Meta are also showing improvements.
Related event: Arena Adds Factuality to Model Rankings(12 posts)→
More from Models
- Rumor claims Kimi faked performance by serving Claude; DeepSeek new model surprises in evals — realsohamparekh · 2026-09-11
- GPT-5.6 writes well but is instantly forgettable, user complains — BasedRaddka · 2026-09-11
- Opus Refuses Protein Research Codebase Over 'Safety' Concerns, Dev Considers Rolling His Own — josephdviviano · 2026-09-11
- User Hails Unconfirmed 'DeepSeek 4.1 Flash' as an Inflection Point in LLMs — himanshustwts · 2026-09-11
- Terminal Bench v4: GLM-5.3 Leads at 41.9%, Kimi-K3 Underwhelms at 12.6% — Ok_Warning2146 · 2026-09-11
- GPT-6 Astra beats Factorio with enemies in 44 in-game hours at ~$4,500 API cost — liminal_bardo · 2026-09-11