DeepSeek
China's frontier open-weight LLM lab — now shipping the V4 family at a fraction of rivals' cost
Overview
DeepSeek is a Chinese AI lab that became globally significant in early 2025, when its open-weight V3 and R1 reasoning models matched far costlier frontier systems at a tiny fraction of the training and inference cost.
In 2026 the lab shipped the DeepSeek V4 family — V4-Pro (a ~1.6-trillion-parameter mixture-of-experts model with roughly 49B active parameters) and the lighter V4-Flash — both with 1-million-token context and strong agentic and coding performance; V4-Pro posts one of the highest open-weights scores on SWE-bench Verified. The models are available free at chat.deepseek.com and through an OpenAI-compatible API. The API uses peak and off-peak billing: usage during peak hours (01:00–04:00 and 06:00–10:00 UTC on weekdays) is full price, while all other times cost half. The openly licensed weights can also be self-hosted, avoiding API costs entirely. (See also our dedicated DeepSeek V4 entry.)
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 31 Aug 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
Use cases
What you can produce with DeepSeek
- Debug and refactor a multi-file codebase with architectural explanations, rivalling paid tools at near-zero API cost
- Solve graduate-level mathematics problems with rigorous step-by-step proofs, including competition-level math (the specialized DeepSeek-Math-V2 model reached gold-medal level at IMO 2025; the default chat model is strong but not IMO-gold)
- Analyse a 300-page document or contract and produce an accurate cited summary, leveraging its long-context processing strength
- Generate and iterate on complete API integration code across multiple programming languages, with web-search grounding to pull in current documentation
- Run local, self-hosted AI inference on your own hardware via the MIT-licensed open weights, enabling air-gapped or privacy-sensitive deployments
- Perform deep research synthesis on complex topics — combining live search, multi-step reasoning, and source citation — at no cost on the free web interface
- Automate data-intensive workflows such as bulk financial or scientific data analysis with Python code generation, debugging, and complexity optimisation
DeepSeek is the Chinese lab whose open-weight models repeatedly shock the market by matching far costlier frontier systems at a fraction of the price. The 2026 V4 family (V4-Pro and the cheaper V4-Flash, both 1M-context) extends that story with top-tier coding and reasoning performance and an extremely cheap, OpenAI-compatible API; the open weights enable self-hosting.
It suits developers and cost-sensitive teams who want strong reasoning on a tight budget, or who must self-host to control data. Caveats remain: the hosted service stores data in China, raising compliance and data-residency concerns for regulated ASEAN/EU use; models reflect Chinese content moderation on sensitive topics; and enterprise support is thin. Self-hosting the open weights sidesteps the residency issue. Capability and value are exceptional.
What people say
Four stars and change is where the crowd lands: the Android app sits around 4.1 from 274,000+ Play Store reviews, and the number has drifted down since a May 2026 update tightened usage limits. The gripe that keeps surfacing is the edit-and-regenerate cap — roughly six goes before the app cuts you off, which turns iterative prompting into rationing. Trustpilot splits between people amazed this much reasoning costs nothing and people tired of 'server busy' errors at peak hours.
The V4 launch (public preview 24 April 2026) drew its own complaints: chat instances that still identified themselves as V3, confusion over which endpoints served which model, and weeks of thin third-party benchmark coverage. Reviewers who tested V4-Pro on real code call it the best value on the market — MakerStack scored it 8.2/10 — while noting it trails GPT-5.5 on front-end polish and shipped with no multimodal support, a real problem for teams whose agents need to see the UI they're building.
Then there's the geopolitics, which for APAC buyers is not a footnote. Prompts, responses and device data are stored on servers in China, subject to the 2017 National Intelligence Law, and researchers at Feroot found code linking the web platform to China Mobile's authentication infrastructure. Italy, Australia, Taiwan, South Korea, India, the Czech Republic and at least 17 US states have banned the app from government devices. Singapore hasn't, but regulated enterprises mostly treat the hosted service as off-limits and self-host the openly licensed weights instead — which is exactly what the open release makes possible.
Summary of public user & expert reviews, compiled by RECATOOLS.
Notable facts
- DeepSeek R1 was released under the MIT licence — the most permissive open-source licence possible — allowing anyone to use it commercially without restrictions.
- The reported training cost of DeepSeek V3 was approximately $5.6 million, compared to estimated hundreds of millions for comparable Western frontier models.
- DeepSeek's app briefly became the #1 downloaded app in the US App Store in January 2025, triggering a 17% drop in NVIDIA's stock price.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including DeepSeek's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with DeepSeek unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to DeepSeek directly →
Spotted something out of date? Suggest an update →
DeepSeek: pricing and development
The cheapest frontier tier, tracked as it moves — endpoint retirements, benchmark claims, and the shift to charging by the clock.
- Everyone Is Cutting Except DeepSeek — and It Now Charges by the Clock
- DeepSeek Retired Its Legacy Endpoints — the Migration Is Free, the Headline Price Is Not
- DeepSeek V4-Pro Takes the Cost-Performance Crown on SWE-bench at 80.6% — 34x Cheaper Output Than GPT...
From our guides
Guide · 9 min read
Why It Confidently Makes Things Up
We asked two AI models for five peer-reviewed studies on an obscure topic, with...
Guide · 16 min read
Self-Host Dify with Docker Compose: 15 Containers to Your Own LLM App Studio (2026)
We deployed Dify 1.16.0 on Docker Desktop, hit a real port clash, wired DeepSeek...
Guide · 9 min read
The Best Open-Source LLMs You Can Actually Self-Host (2026)
Qwen, Llama, DeepSeek, Mistral, Phi and Gemma compared for self-hosting in 2026...
More in LLMs & Chat
