DeepSeek V4.1-Flash-1017: MIT weights, 77% DeepSWE, ~$0.20 per task
A breakdown of the release: MIT open weights, shorter outputs cutting agent cost and latency, 77%+ on DeepSWE at about $0.20 per task versus $0.50 for the prior Flash, plus Search/Fetch APIs at $0.75 per 1k calls.
DeepSeek shipped V4.1-Flash-1017 today — MIT open weights, shorter outputs, faster agent runs. On their DeepSWE chart it clears 77% at ~$0.20/task (prior Flash ~$0.50). Vendor benches also put it ahead of Opus 5 and GPT-5.6 Sol on several coding/agent suites — wait for third-party before you rewrite the stack. The real ops story is the shorter outputs. Agent fleets burn money and time on verbose loops; a model that finishes with less text is a cost and latency lever, not just a benchmark bump. They also opened Search/Fetch APIs at $0.75/1k calls. Agents that can fetch will fetch a lot — budget the egress. https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017