DeepSeek V4.1 Flash Promises Cheaper AI Coding at $0.10 a Task
Chinese AI company DeepSeek has released DeepSeek V4.1 Flash, an updated version of its Flash model (V4.1-Flash-1017) with improvements in coding, agentic tasks, speed and output efficiency — and a price point that undercuts far larger rivals.
The model is available through DeepSeek’s website, app and API. The company has also published its weights on Hugging Face and ModelScope, keeping the open-weights approach that made its earlier releases popular with developers, ProPakistani reported. DeepSeek V4.1 Flash is aimed squarely at end-to-end software development, from game development and 3D modelling to design work and even compact animated explainer videos.
DeepSeek V4.1 Flash uses a new training method
For this release, DeepSeek used a post-training method called reinforcement learning with self-feedback, or RLSF. In plain terms, the model grades its own work — using its own confidence scores or generated critiques as reward signals — instead of relying on human evaluators or external reward models. DeepSeek says this improves the model’s ability to build complete projects from scratch.
The company tuned V4.1-Flash-1017 for development environments including DeepSeek Harness, Tencent CodeBuddy and WorkBuddy, and recommends DeepSeek Harness version v0.2.1-rc.1 or newer.
Cheaper than the big names
According to DeepSeek’s own benchmarks, V4.1-Flash-1017 improves over its predecessor across every tested category. The company also claims the smaller model beats larger systems including Claude Opus 5, GPT-5.6 Sol, Kimi K3 and GLM 5.3 on several coding and AI agent benchmarks.
The headline numbers are about cost. On the DeepSWE benchmark, V4.1-Flash-1017 scored 77.0% at an average cost of just under $0.20 per task. At its low setting it still scored above 72% while costing around $0.10 per task. Its predecessor managed 74.2% but cost roughly $0.50 per run. DeepSeek’s chart places models such as GPT-6 Astra, Gemini 3.8 Flash and Claude Opus 5 at between roughly $2 and more than $10 per task, with lower accuracy.
Two caveats matter. First, independent benchmarks are not yet available, and real-world developer assessments of code quality may differ from benchmark results. Second, the release comes as the Chinese government reportedly considers tighter controls on publishing open-weight AI models — a move that could complicate deeper reliance on Chinese open models if export or publication rules tighten.
What it means for Pakistani developers
The cost angle is the real story for Pakistan. At around $0.10 per task — roughly Rs28 at current rates — automated coding agents become affordable for freelancers, small software houses and startups that could never justify $2 to $10 per task on premium models. Cheaper agentic coding also matters for the kind of AI-assisted workflows Pakistani teams are already experimenting with, as the country’s national AI programme seeks World Bank financing.
The AI race itself keeps accelerating on other fronts too — from Anthropic drafting its own AI safety law for US lawmakers to Google pulling Pro models from its AI Plus tier, and even North Korea claiming an AI-guided torpedo test.
Search APIs opened to developers
Alongside the model, DeepSeek is opening its internal web search technology to outside developers for the first time. The new Search and Fetch APIs use DeepSeek’s own search engine, which the company says already handles hundreds of millions of queries per day worldwide. Each endpoint costs $0.75 per 1,000 calls.
DeepSeek API users can access the tools free through server-side integrations in DeepSeek Harness, Claude Code and Codex until October 31, 2026, at 4:00 PM UTC.
Frequently asked questions
What is DeepSeek V4.1 Flash?
It is an updated version of DeepSeek’s Flash AI model (V4.1-Flash-1017), focused on coding and agentic tasks, with open weights published on Hugging Face and ModelScope.
How much does it cost per task?
DeepSeek claims about $0.20 per task at top settings on the DeepSWE benchmark, and around $0.10 per task at its low setting — far below the $2 to $10+ per task it attributes to larger rival models.
What is RLSF training?
Reinforcement learning with self-feedback: the model uses its own confidence scores or generated critiques as reward signals during training, rather than relying on human evaluators.
Are the benchmark claims verified?
Not yet by independent testers. DeepSeek’s figures are its own; independent benchmarks are not available, and real-world developer experience may differ.
