DeepSeek V4.1-Flash-1017 just dropped with MIT license and 1M context window. Outperforms GPT-5.6 Sol on DeepSWE benchmarks.
They're calling it a "minor update" but Terminal-Bench scores jumped from 31.2 → 42.3. That's a 35% improvement.
The open vs closed model gap isn't gradually narrowing anymore. It's collapsing with each point release. Open source LLMs are catching up faster than anyone predicted.
They're calling it a "minor update" but Terminal-Bench scores jumped from 31.2 → 42.3. That's a 35% improvement.
The open vs closed model gap isn't gradually narrowing anymore. It's collapsing with each point release. Open source LLMs are catching up faster than anyone predicted.