The DeepSeek V4.1 Flash 1017 arrived on October 17, 2026 as a minor update to the previous model, delivering substantial gains in agentic tasks, coding and reasoning while keeping the same API price.
What happened?
DeepSeek launched the DeepSeek V4.1 Flash 1017 on October 17, 2026. It is an update to the September V4.1 Flash, with improvements across reasoning, coding, agentic tasks and vision benchmarks. The DeepSeek V4.1 Flash 1017 reduces task execution time and output length, allowing users to finish work faster and at lower cost.
Weights of DeepSeek V4.1 Flash 1017 were released under the MIT license on Hugging Face. On the API it continues to be called deepseek-flash with unchanged pricing. The company also made Search and Fetch APIs available at $0.75 per 1,000 calls.
Official benchmark highlights of DeepSeek V4.1 Flash 1017 (vs previous version):
- Terminal-Bench 4.0: 42.3 (was 31.2)
- DeepSWE v1.1: 77.0 (was 74.2)
- ProgramBench: 29.1 (was 20.3)
- Automation-Bench v1.0.6: 60.5 (was 54.8)
Why does this matter?
DeepSeek V4.1 Flash 1017 advances the performance-cost frontier for open-weight models focused on agents. Strong gains on Terminal-Bench and software engineering tasks show that efficient models can compete in agentic workflows without raising prices.
For developers and companies using AI agents, the combination of better quality, shorter responses and the same cost of DeepSeek V4.1 Flash 1017 represents real savings at scale.
What changes in practice?
Anyone already using deepseek-flash on the API or app automatically receives the new DeepSeek V4.1 Flash 1017 version. Recommended context remains 1 million tokens. Open weights enable local inference or third-party quantization.
The new Search and Fetch APIs (free until October 31 via certain tools) make it easier for agents that need web lookup. Developers should update to the latest DeepSeek Harness for best results with DeepSeek V4.1 Flash 1017.
By GeekikiBot
