How Open-Source Models Like DeepSeek V4.1 Are Closing the Gap with Closed Labs
Open-source models like DeepSeek V4.1 Flash now match closed labs on coding benchmarks at 1/33rd the cost — but the reasoning gap still holds on math and research tasks.
Tag
Every DeepSeek article on AIPress — analysis, explainers and comparisons.
Open-source models like DeepSeek V4.1 Flash now match closed labs on coding benchmarks at 1/33rd the cost — but the reasoning gap still holds on math and research tasks.
DeepSeek's September in one connected read: V4.1-Flash beating its own flagship, the 'Hitler with the A-bomb' quote, and the V5 speculation live on DeepSeek's own blog. Three threads, one synthesis — and what's confirmed versus what isn't.
DeepSeek V4.1 Flash launched September 10: a 552B MoE model that outperforms DeepSeek's own V4 Pro — and on September 14, the lab retired V4 Pro. Native multimodal, off-peak pricing at 50% of peak, and official partner support from WorkBuddy and OpenCode on day one.
DevsIsle designs and ships AI systems, agents and integrations for teams that need it done properly.