Deepseek
DeepSeek's chat stays free in 2026, while its V4 API adds peak/off-peak pricing and a 1M+ token context window via Engram memory technology.
About Deepseek
What is DeepSeek (2026 update)
DeepSeek is a Chinese AI research company that builds open-weight foundation models and offers both a free consumer chatbot and a pay-per-token API for developers. It became known for undercutting Western AI labs on price while publishing competitive benchmark results, and by September 2026 its main product is the V4 model family.
What is new in 2026
DeepSeek V4 launched on April 24, 2026, with open weights available across chat, mobile app, and API, split into two variants: V4-Flash for speed and cost efficiency, and V4-Pro for heavier workloads. V4 integrates what DeepSeek calls Engram conditional memory technology, which lets the model retrieve information efficiently from contexts exceeding one million tokens. DeepSeek’s own internal benchmarks claim V4 outperforms Claude and GPT-series models specifically on long-context code generation tasks, though these are self-reported and worth verifying independently. On the pricing side, DeepSeek introduced peak and off-peak API pricing on August 16, 2026, charging more during UTC peak hours (01:00-04:00 and 06:00-10:00) than during the rest of the day. A widely rumored reasoning model, R2, has not shipped as of mid-2026; reports indicate DeepSeek’s CEO delayed its release over internal quality concerns, so R2 remains unannounced.
Key features
- Open-weight V4 models (V4-Flash and V4-Pro) usable via chat, app, or API
- Engram conditional memory technology for efficient retrieval across contexts over 1 million tokens
- Free, unrestricted consumer chat with no Plus or Pro paywall
- Pay-per-token API pricing with a free onboarding token grant for new developers
- Peak/off-peak API pricing that rewards running jobs during off-peak UTC hours
- Prompt caching that cuts input token cost sharply on repeated context
Pricing in 2026
DeepSeek’s consumer chat has no subscription tier; only the API is metered.
| Plan | Price | What you get |
|---|---|---|
| Web and app chat | Free | Full access to DeepSeek’s chat interface with no message paywall, Plus plan, or Pro subscription |
| Developer free grant | Free | New API accounts get 5 million tokens valid for 30 days, no credit card required, worth roughly $8-$10 at flat V4-Flash rates |
| V4-Flash API (off-peak) | $0.22 input / $0.66 output per 1M tokens | Standard hours outside the peak window; cache hits cut input cost to about $0.007 per 1M tokens |
| V4-Flash API (peak) | $0.44 input / $1.32 output per 1M tokens | Applies during UTC peak hours (01:00-04:00 and 06:00-10:00) |
| V4-Pro API | $0.66 input / $1.98 output per 1M tokens | Higher-capability model for heavier or more complex API workloads |
Who should use it
DeepSeek fits developers and businesses who need a low-cost API for high-volume tasks, especially long-context work like large codebase analysis, where Engram memory and V4’s long-context claims matter most. Casual users who want a capable free chatbot without any subscription also get full functionality at no cost. Teams sensitive to API cost can schedule batch jobs during off-peak UTC hours to cut spend further.
Limitations and alternatives
DeepSeek’s benchmark claims against Claude and GPT models come from its own internal testing, so independent verification is advisable before relying on them for a purchasing decision. The peak/off-peak pricing model, introduced in August 2026, adds complexity that flat-rate competitors do not have, and R2, the expected next reasoning model, has no confirmed release date, so check the official site for the latest status. As a China-based company, some enterprises may have data residency or compliance concerns that rule DeepSeek out for regulated workloads. For free general-purpose chat, ChatGPT’s Free tier and Google’s Gemini are direct alternatives. For lower-cost API access with different long-context tradeoffs, Google’s Gemini Flash line and OpenAI’s smaller models are commonly compared against DeepSeek’s V4-Flash pricing.
Frequently asked questions
Is DeepSeek free to use?
Yes, DeepSeek's web and app chat is completely free with no Plus or Pro subscription. Only the developer API is metered per token.
What is DeepSeek V4?
V4 is DeepSeek's model family launched April 24, 2026, split into V4-Flash and V4-Pro, featuring Engram conditional memory technology for efficient retrieval across contexts exceeding one million tokens.
Has DeepSeek released R2?
No, as of the most recent reports R2 has not been officially released; DeepSeek's CEO reportedly delayed it over internal quality concerns, so check the official DeepSeek blog for the current status.
Why does DeepSeek's API pricing change during the day?
Since August 16, 2026, DeepSeek charges higher per-token rates during UTC peak hours (01:00-04:00 and 06:00-10:00) and lower rates the rest of the day, so scheduling jobs off-peak reduces cost.
Information last verified: September 2026.
Related Tools
