DeepSeek releases open V4.1-Flash-1017 under MIT license and opens paid Search and Fetch APIs

The Hugging Face model card describes V4.1-Flash-1017 as a minor update to V4.1-Flash with a 1M-token context window and MIT-licensed weights. DeepSeek's own table lists 77.0 on DeepSWE v1.1, ahead of Opus 5 at 74.0 and GPT-5.6 Sol at 73.0, and 42.3 on Terminal-Bench 4.0 against 51.8 for Opus 5. Opus 5 still leads on HLE (56.3 versus 42.6) and several other rows, and all figures are self-reported. On the API side, the new Search and Fetch endpoints bill about $0.75 per 1,000 successful requests each, with a limit of 5 requests per second per account. Secondary coverage of DeepSeek's docs says server-side calls made through DeepSeek Harness, Claude Code and Codex are free until October 31. The model also now serves the deepseek-flash API name, so existing integrations will change behavior. For builders, an open agent model paired with a first-party retrieval API lowers the cost of building research and coding agents without a separate search vendor. Teams should rerun their own evals before switching, since the benchmarks are vendor-reported and the free window ends in two weeks.