Thirty-two and a half against forty-six. That is the gap on the Artificial Analysis Intelligence Index v5 between DeepSeek-V4.1-Flash-1017 and Moonshot AI’s Kimi K3.1 Pro, and it is the first thing to know about DeepSeek’s latest release. The second is the price tag, which is tiny.
The shift is toward cheap models that are good enough, and DeepSeek is pushing hardest. Its chat app is free with no paid tier. The API charges $0.15 per million input tokens and $0.60 per million output tokens off-peak, rising to $0.30 and $1.20 at peak.
Who is behind it
DeepSeek is a Hangzhou company funded by the Chinese hedge fund High-Flyer and founded by Liang Wenfeng. It releases open-weight models, and the 1017 build, which landed on October 17, 2026, is an incremental refresh of V4.1-Flash. DeepSeek says it improves reasoning, coding, agentic work and visual benchmarks.
Architecture
V4.1-Flash is a multimodal Mixture-of-Experts model with 552 billion backbone parameters and a context window of up to one million tokens. It ships under the MIT license, about as permissive as software licensing gets. Anyone may modify and redistribute it, and people did. “Abliterated” and “uncensored” forks of the base model showed up on Hugging Face within hours of its release.
Cost and speed
What’s changing is how little a decent model costs to run. Artificial Analysis clocked 173 tokens a second on DeepSeek’s own API, against 41 for Kimi K3.1 Pro. The October update also brought built-in web search for the first time: Search and Fetch calls cost $0.75 per 1,000 successful requests, with a limit of 5 requests per second per account.
For developers
There is no Jinja chat template in this release. Instead, a self-contained Python reference implementation covers multi-turn chats, tool calling and thinking mode. For production work, DeepSeek offers deepseek-recipe, Rust libraries with Python bindings that turn API requests into the Conversation format.
Watch this space on November 4. That is the date Moonshot has promised to publish the Kimi K3.1 Pro weights, which would put the stronger model in the open-weight race too.
Sources
- deepseek-ai/DeepSeek-V4.1-Flash-1017 · Hugging Face
- DeepSeek
- DeepSeek V4.1 Flash Uncensored: 2,254 Downloads [2026]
- Kimi vs DeepSeek (2026): Which Chinese AI Should You Use?
- Activities for This Week and Next from AI News: 2026/10/30 Edition｜おーら
- GPT-6 Astra Distillation: DeepSeek, Moonshot Named in Report
Photo: Štefan Štefančík / Unsplash (Unsplash License)
