DeepSeek icon

DeepSeek

DeepSeek-V4-Flash-0731 Public Beta

Delivers open-weight V4 frontier models (Pro 1.6T / Flash 284B) with 1M context, MIT license, and OpenAI/Anthropic API compatibility.

Reviewed by ToolWorthy Editors·updated 3 months ago·DeepSeek-V4-Flash-0731 Public Beta released 11 days ago

Pricing:From ¥0.02/per M cache-hit input tokens
Categories:
Jump to section
DeepSeek sign-in page for the official web chat assistant

More tools to compare

Omniwork icon

Omniwork

Coasty icon

Coasty

MakersClaw icon

MakersClaw

Blocks.ai icon

Blocks.ai

BrowserAct icon

BrowserAct

YAGNI icon

YAGNI

Pros & Cons

Pros

  • Official web chat is free
  • Strong bridge from end-user chat to API usage
  • Clear public product structure across chat, app, and API
  • Useful for users who want current DeepSeek models without immediate payment
  • Research lineage strengthens product credibility

Cons

  • Web chat requires sign-in
  • API pricing is a separate layer from the free assistant
  • Product differences between app, web, and API can confuse first-time users
  • Some enterprise governance needs may require more evaluation
  • Public chat access does not equal predictable API cost

Overview

DeepSeek is the public-facing AI assistant and model platform from DeepSeek. The official web experience is the chat app at chat.deepseek.com, while the company’s main site also points users toward the app, the API platform, and model-specific research releases.

The key distinction is that DeepSeek’s consumer chat and its API are not the same pricing layer. The official site explicitly promotes free web dialogue with the current flagship model, while API usage sits behind separate token-based pricing on the platform side.

As of April 26, 2026, DeepSeek’s official site says users can start a free conversation with DeepSeek-V3.2 on the web and app. The API pricing documentation separately describes paid token pricing for model access. That makes the product attractive both to end users who want a free assistant and to developers evaluating the models programmatically.

For adjacent research, compare AI design tools, AI UI design tools, AI website design tools.

Key Features

  • Free official web chat - DeepSeek’s product site explicitly promotes free conversation in the web version.

  • Current flagship model access - The official site ties the chat experience to DeepSeek-V3.2.

  • Separate API platform - DeepSeek also offers a dedicated platform for model access beyond the consumer web chat.

  • Research-to-product pipeline - The product site links out to DeepSeek’s research families, including V3, R1, Coder, and multimodal work.

  • Simple web-first onboarding - The chat experience is accessible from the browser with sign-in, keeping friction relatively low for end users.

  • Developer expansion path - Users who outgrow the free chat can move to the API stack for structured integration and paid usage.

Pricing & Plans

DeepSeek’s public pricing story splits into two parts: free end-user chat and paid API consumption.

Option Pricing Best fit
DeepSeek web chat Free Users who want direct conversation with the official assistant in the browser
DeepSeek app Free to access the assistant experience Mobile users who want the same assistant on mobile
DeepSeek API Token-based paid usage Developers integrating models into apps and workflows

The official DeepSeek site currently says users can start a free conversation with DeepSeek-V3.2. The API pricing docs separately note that API pricing is billed by token usage and that it differs from the web and app experience. For most users, the practical question is whether the free web assistant is enough or whether they need programmable API access.

Best For

  • Users wanting a free official AI assistant in the browser
  • Developers evaluating DeepSeek before integrating the API
  • People comparing DeepSeek’s web assistant to other frontier chat tools
  • Teams following the DeepSeek research and product roadmap
  • Users who may later move from chat to API workflows

FAQ

Is DeepSeek free?

The official site says the web and app assistant experience is free to start for conversation with the current flagship model.

Does DeepSeek have paid pricing?

Yes. Paid pricing applies on the API side, where usage is billed separately from the free web chat.

What is the difference between DeepSeek Chat and the API?

DeepSeek Chat is the end-user web assistant, while the API is for developers integrating models programmatically with token-based billing.

Which model does the official site promote?

The current site promotes free conversation with DeepSeek-V3.2.

Who should use DeepSeek?

It is a good fit for users who want a free official assistant now and developers who may later need API access to the same model family.

What is DeepSeek?

DeepSeek is an AI tool reviewed for its core features, pricing, strengths, limitations, and best-fit workflows.

Version History

DeepSeek-V4-Flash-0731 Public Beta

Released on July 31, 2026

View Update
+What's new
3 updates
  • Use the formal public-beta DeepSeek-V4-Flash API through the same deepseek-v4-flash model name with no request-format change
  • Evaluate DeepSeek-reported Code Agent scores across Terminal Bench 2.1, NL2Repo, Cybergym, DeepSWE, Toolathlon, and DSBench; no Flash Preview baseline is disclosed
  • Build Codex-style agents with native Responses API support while V4-Pro, DeepSeek APP, and DeepSeek WEB models remain unchanged for now

DeepSeek-V4 Preview

Released on April 24, 2026

View Update
+What's new
3 updates
  • Run million-token coding, reasoning, and agent workflows on DeepSeek-V4-Pro (1.6T/49B active MoE) or faster V4-Flash (284B/13B active MoE)
  • Use V4-Flash at $0.14/M cache-miss input and $0.28/M output, or V4-Pro at $1.74/M cache-miss input and $3.48/M output; open weights are MIT licensed
  • Compare against frontier models with V4-Pro-Max scoring 80.6% on SWE Verified, 67.9 on Terminal Bench 2.0, and leading current open models on knowledge and reasoning

DeepSeek-V3.2

Released on December 1, 2025

+What's new
3 updates
  • Integrate thinking directly into tool use with DeepSeek-V3.2, the first model to support tool-use in both thinking and non-thinking modes
  • Use DeepSeek-V3.2 as the official successor to V3.2-Exp across App, Web, and API, with both thinking and non-thinking tool-use modes
  • Launch DeepSeek-V3.2-Speciale (API-only) alongside, pushing reasoning capabilities further for the most demanding research workloads

DeepSeek-V3.2-Exp

Released on September 29, 2025

+What's new
3 updates
  • Explore DeepSeek Sparse Attention (DSA) for the first time, delivering fine-grained sparse attention that dramatically improves long-context training and inference efficiency
  • Cut API prices by more than 50% versus V3.1-Terminus while keeping benchmark performance broadly comparable across most evaluated tasks
  • Validate architectural optimizations for long-context scenarios on open weights before rolling DSA into the December V3.2 general release

DeepSeek-V3.1-Terminus

Released on September 22, 2025

+What's new
3 updates
  • Get cleaner language consistency across extended conversations, reducing mid-response switching between English and Chinese that appeared in earlier V3.1 output
  • Use stronger Code Agent and Search Agent behaviors with more reliable tool invocation for coding and multi-step web research workflows
  • Refine the August V3.1 experience one month later with cleaner language consistency and stronger agents, preparing the path to V3.2-Exp

DeepSeek-R1

Released on January 20, 2025

+What's new
3 updates
  • Solve advanced reasoning tasks on math, science, and coding with DeepSeek-R1's reinforcement-learning approach, rivaling OpenAI o1 at a fraction of the cost
  • Score 79.8% pass@1 on AIME 2024 and 97.3% on MATH-500 while running as open weights under MIT license—rare for frontier reasoning models at the time
  • Use DeepSeek-R1 and six distilled checkpoints for local reasoning workflows, with the main model under MIT and each distilled model retaining its base-model license

DeepSeek-V3

Released on December 26, 2024

+What's new
3 updates
  • Launch DeepSeek-V3 as a 671B-parameter MoE model with 37B activated per forward pass and 128K context, outperforming most open models on coding and math
  • Cut training cost to roughly $5.6M by combining MoE routing with FP8 mixed-precision training, dramatically undercutting Western frontier model economics
  • Ship open weights for both base and chat variants with full tokenizer and inference code, setting a new bar for open-source frontier model transparency

DeepSeek-V2

Released on May 6, 2024

+What's new
3 updates
  • Release DeepSeek-V2 with a 236B-parameter MoE architecture and 21B activated per token, delivering stronger performance at lower inference cost than dense peers
  • Use substantially lower-cost API inference and access DeepSeek-Coder-V2 from June 17, 2024 for dedicated coding and mathematical reasoning workloads
  • Establish the Mixture-of-Experts plus low-price playbook that defines DeepSeek's strategy through the V3, R1, and V4 generations

DeepSeek LLM

Released on November 29, 2023

+What's new
2 updates
  • Download DeepSeek LLM 7B or 67B base and chat checkpoints as DeepSeek's first general-purpose LLM release, trained on two trillion English and Chinese tokens
  • Use the released tokenizer, training details, intermediate checkpoints, and chat variants; code is MIT-licensed while model use follows DeepSeek's commercial-use model license

Top alternatives

Related categories

From the blog

View all →

Track DeepSeek in ToolWorthy Weekly

Important tool updates, better alternatives, and selected AI signals in one weekly brief.

Weekly only. Unsubscribe anytime.