By Hamza Faisal

DeepSeek is not slowing down. The Chinese AI lab has released DeepSeek-V4.1-Flash-1017, an updated checkpoint of the V4.1-Flash model it launched in September — and it is framing the release as a straight win on both performance and cost.

The new model is live now on DeepSeek's website, app and API, served under the existing deepseek-flash model name, so developers get it without changing their integration or paying higher rates. The weights are also open-sourced under the MIT license on Hugging Face and ModelScope.

Better scores, same price

DeepSeek calls the update "a Pareto improvement in performance and cost": stronger benchmark scores with faster task completion and shorter outputs. On its own comparison tables, the model scores 77.0 on the DeepSWE v1.1 coding benchmark, up from 74.2 — ahead of Claude Opus 5 (74.0) and GPT-5.6 Sol (73.0).

The gains extend across the board. Terminal-Bench 4.0 jumps from 31.2 to 42.3, ProgramBench climbs from 20.3 to 29.1, and DeepSeek says the new checkpoint beats the previous Flash on all 11 benchmarks it reported, by 2.0 to 11.1 points. One caveat is worth noting: these are company-reported numbers, not independent tests — and on raw reasoning without tools, Claude Opus 5 still leads.

The model is built for agentic work. DeepSeek says it used reinforcement learning with self-feedback to train the model to build complete projects end to end — from games to 3D models — and tuned it for its own DeepSeek Harness as well as Tencent's CodeBuddy and WorkBuddy.

A search engine of its own

The model was not the only launch. DeepSeek also opened its Search and Fetch APIs to all developers, powered by an in-house search engine it says serves hundreds of millions of requests a day.

For agent builders, the combination matters: a low-cost coding model plus a first-party web layer behind a single API key. Each search endpoint costs $0.75 per 1,000 calls, and calls routed through server-side tools in DeepSeek Harness, Claude Code and Codex stay free until per-request billing starts on November 1.

That puts DeepSeek in direct competition with dedicated search-API providers for agent workloads — a market that has been quietly consolidating as every agent framework adds web access.

The cost war keeps escalating

The economics are the real story. DeepSeek's own API pricing sits at $0.15 per million input tokens and $0.60 per million output tokens — and an independent 56-trial pilot cited this week put a completed task at $0.0157 on Flash 1017, versus $0.0577 for GPT-6.1 Sol and $0.0653 for Claude Sonnet 5.5.

That gap is why the release matters beyond the benchmark tables. If agents are going to do real work at scale, somebody has to make the tokens cheap — and DeepSeek keeps volunteering.

The release lands as DeepSeek's latest funding round has reportedly grown to between $12 billion and $15 billion at a valuation of about $75 billion, ahead of a planned Shanghai IPO, according to CNBC. The lab is spending like a company that expects its models to carry the business, not just the headlines.

By Hamza Faisal