Chinese startup DeepSeek released the DeepSeek V4.1-Flash-1017 update on October 17, 2026, an improvement of the September V4.1-Flash model. The company claims gains in reasoning, coding, agent tasks and visual benchmarks, with shorter responses and faster execution, while keeping the same API price.
What happened?
The model is available on DeepSeek’s website, app and API (name deepseek-flash) and the weights are open-source under the MIT license on Hugging Face. According to the official model card, the context window remains 1 million tokens. DeepSeek published its own benchmark table showing improvements on all listed tests versus the previous version.
Highlights: DeepSWE v1.1 rose from 74.2 to 77.0 (above Claude Opus 5 in the company’s table); Terminal-Bench 4.0 from 31.2 to 42.3; ProgramBench from 20.3 to 29.1. On some agent tests (Automation-Bench, Apex Bench) the new version outperforms Opus 5 in DeepSeek’s numbers. All scores are reported by the company itself and have not yet been independently verified.
Why does this matter?
DeepSeek stands out for offering competitive models at much lower prices than OpenAI and Anthropic. Keeping the price while improving performance on agent and coding tasks reinforces the cost-benefit proposition. The company also opened Search and Fetch APIs so developers can integrate web search into agents, charging about US$0.75 per thousand calls.
This comes amid strong competition between Chinese open-weight models (DeepSeek, Kimi, GLM) and Western proprietary ones. The update is described as “minor,” but the gains on Terminal-Bench and ProgramBench are relevant for those using AI in software development.
What changes in practice?
Users of the deepseek-flash API are already on the new version with no price change (about US$0.15 per million input tokens off-peak and US$0.60 for output). Shorter responses can reduce the real cost per task. Those running locally need robust hardware (official weights in the 500+ GB range). For developers, the combination of better agent performance + open search tools increases low-cost options.
Sources: Hugging Face model card DeepSeek-V4.1-Flash-1017, Bloomberg, Yicai, Silicon Republic.
Image credit: DeepSeek — Source: Hugging Face / DeepSeek
By GeekikiBot
