DeepSeek v4.1 flash paper figure shows sharp quality jump at 1M-token context — agents are context-hungry
TheZachMueller · x · 2026-09-19
andrewncarr highlighted a must-study figure from the DeepSeek v4.1 flash paper: quality rises sharply once context length is extended to 1M tokens, with similar gains visible in MiMo v2.6's RL graphs. His takeaway: agents are context-hungry. Replying, Zach Mueller joked, "brb as I load all of Wikipedia into context every turn."
More from Models
- Dev Reviews 287 Open-Source Jev Projects, Picks 20 That Show What It's Good At — chenrongwei · 2026-09-19
- Qwen3.8-Omni-Flash undercuts Gemini Flash pricing while matching its multimodal benchmarks — The Decoder · 2026-09-19
- GLM 5.3F tested: good at Remotion, fails at complex JS-coded videos — 9r4n4y · 2026-09-19
- JevBench v1.2 released: Jev 1.13 holds the lead at 75.3, open 4B models close behind — airesearch12 · 2026-09-19
- Fine print: $3,800 per task and still no paradigm-shifting breakthroughs over humans — seanwbren · 2026-09-19
- Jev fills the general-purpose classifier gap: labels in code, no extra LLM call — alexcovo_eth · 2026-09-19