DeepSeek-V4-Flash Lands on Agent Arena at #3 Among Open-Source Models
infwinston · x · 2026-08-04
DeepSeek officially launched the DeepSeek-V4-Flash API in public beta, massively upgrading its Agent capabilities with native support for the Responses API and Codex.
Based on 12.5K real-world agentic sessions in Agent Arena, the model achieved the following:
- Overall Ranking: Landed at #21 globally (+1.98% net improvement) and #3 among open-source models.
- Internal Comparison: Ranked 6 spots higher than DeepSeek-V4-Pro and 13 spots higher than the previous V4-Flash.
- Key Signals: Showed strength in Confirmed Success (+6.99%) but was slightly weaker in Steerability (-2.31%).
More from Models
- Gavin Baker said SSI would release its model in August — one week left to verify — imjustnewatai · 2026-08-25
- User Feedback: Claude Max Reasoning Mode Delivers Better Quality with Manageable Usage — CtrlAltDwayne · 2026-08-25
- User says Claude limit fix didn't work: 36% of weekly quota gone in under 24h — StewartalsopIII · 2026-08-25
- GLiNER 2.5 Launches with Architecture Upgrade for Long-Context Extraction — huggingface · 2026-08-25
- Analysis confirms stealth/ox-alpha is a Z.ai GLM model — PawelHuryn · 2026-08-25
- Users complain about Claude's verbose and hollow output style — asyhler · 2026-08-25