DeepSeek news and release timeline
Independent, source-linked timeline of DeepSeek model releases, API and price changes, dsh releases, and papers. Independent resource, not affiliated with DeepSeek.
What is the latest DeepSeek model? As of 17 Oct 2026, the API change log documents DeepSeek-V4.1-Flash-1017 Update (API name deepseek-flash for V4.1 Flash per the 2026-09-10 entry).
What changed most recently in the API? The newest timeline entry (@deepseek-ai/dsh 0.2.2-alpha.3 published on npm, 3 Nov 2026) summarizes the latest official change; see the full timeline below for model retirements, routing, and pricing notes.
Is DeepSeek V4-Pro still available? Per the 2026-09-10 API change log, V4 Pro API service continues after 14 Sep 2026 with billing unchanged until DeepSeek gives further notice.
Last checked 3 Nov 2026. Unofficial independent guide — not affiliated with DeepSeek.
Latest updates
@deepseek-ai/dsh 0.2.2-alpha.3 published on npm
npm dist-tags for @deepseek-ai/dsh list alpha as 0.2.2-alpha.3, published 3 Nov 2026 07:48 UTC, while latest and next remain 0.2.1-rc.5. This is an alpha pre-release version, not a stable GA release. GitHub marks dsh-v0.2.2-alpha.3 as a pre-release. The release notes unify mouse and keyboard context menus across Sessions, workspaces, and the sidebar; keep a directory tree in file previews; unify code review and worktree access in the experimental change-request and worktree feature (searchable file tree and per-file navigation for local and PR diffs, quote selected code and comments to the Agent, track checks and reviews), folding the standalone Git worktree plugin into it (users who enabled @deepseek-ai/dsh-experimental-tool-worktree must remove it from dsh.profile.bundles); move automation tasks to unified host-level scheduled-task management with manual creation, direct editing, and session-based creation; add a new apply_patch tool (off by default); extend TypeScript and Python SDKs with native commands and plugin management; turn read-before-write and external-change checks off by default while keeping sandbox limits and Review; and include several bug fixes.
Official source: https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.2-alpha.3
Related pages: /cli · /cli/install
DeepSeek Harness Creator Mode showcased
On 2 Nov 2026 DeepSeek announced and showcased Creator Mode in DeepSeek Harness (dsh). In Creator Mode you describe a plugin in plain language, select the Creator Mode option, and the agent builds the plugin for use inside Harness. DeepSeek published a prompt collection for the seven demo plugins from its announcement video: 3D gallery of famous landmarks, Transformer 3D visualization, DSH pixelated interface, Poster design workbench, Video editing workbench, Paper reader, and Conversation statistics. DeepSeek notes some demos were refined over several rounds of conversation, so one prompt may not reproduce the video exactly.
Official source: https://cdn.deepseek.com/harness/dsh-creator-mode-prompt-collection-en.html
Related pages: /cli · /cli/install
dsh-superagent-prototype open-sourced
DeepSeek's Harness team open-sourced dsh-superagent-prototype under the MIT licence (repository description: 'Towards a More Autonomous Agent System'; first commit 31 Oct 2026 12:05 UTC, message 'feat: add the dsh-superagent research prototype'). Per its README, this is a research prototype for reference only, not an official DeepSeek product. It is a proactive coding agent for one GitHub repository, packaged as a bundle for DeepSeek Harness (dsh). It listens to GitHub and optionally Feishu; when an allowed person asks for work it opens a work item, works in its own agent session and directory, and answers in the original conversation (for example with a pull request). A web panel shows what needs a person. Its stated principle is 'take as little human attention as possible, and give as much value as possible'. It accompanies a DeepSeek Harness article 'Exploring Super-agent Design: Towards a More Autonomous Agent System', also posted by @DeepSeekHarness on X on 1 Nov 2026. The published code differs from DeepSeek's internal version and some designs in the article are not implemented.
Official source: https://github.com/deepseek-ai/dsh-superagent-prototype
Related pages: /cli · /cli/install
@deepseek-ai/dsh 0.2.1-rc.5 published on npm
npm dist-tags for @deepseek-ai/dsh list both latest and next as 0.2.1-rc.5, published 29 Oct 2026, while alpha remains 0.2.2-alpha.2 (previously latest pointed to 0.2.1-rc.2). This is a pre-release version, not a stable GA release. GitHub marks dsh-v0.2.1-rc.5 as a pre-release, published 30 Oct 2026. The release notes only improve how thinking content is displayed in conversations.
Official source: https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.1-rc.5
Related pages: /cli · /cli/install
@UnslothAI shared new GGUF quantisations of DeepSeek-V4.1-Flash-1017
@UnslothAI shared new GGUF quantisations of DeepSeek-V4.1-Flash-1017 for running the model locally. Unsloth says they run on machines with 64GB+ RAM and reports that its Dynamic quants score highest on benchmarks such as GPQA Diamond, AIME25 and LiveCodeBench v6.
Shared by @UnslothAI (original post)
Related pages: /models/deepseek-v4-1-flash
Current models at a glance
Full timeline
2026
@deepseek-ai/dsh 0.2.2-alpha.3 published on npm
npm dist-tags for @deepseek-ai/dsh list alpha as 0.2.2-alpha.3, published 3 Nov 2026 07:48 UTC, while latest and next remain 0.2.1-rc.5. This is an alpha pre-release version, not a stable GA release. GitHub marks dsh-v0.2.2-alpha.3 as a pre-release. The release notes unify mouse and keyboard context menus across Sessions, workspaces, and the sidebar; keep a directory tree in file previews; unify code review and worktree access in the experimental change-request and worktree feature (searchable file tree and per-file navigation for local and PR diffs, quote selected code and comments to the Agent, track checks and reviews), folding the standalone Git worktree plugin into it (users who enabled @deepseek-ai/dsh-experimental-tool-worktree must remove it from dsh.profile.bundles); move automation tasks to unified host-level scheduled-task management with manual creation, direct editing, and session-based creation; add a new apply_patch tool (off by default); extend TypeScript and Python SDKs with native commands and plugin management; turn read-before-write and external-change checks off by default while keeping sandbox limits and Review; and include several bug fixes.
What it means: Install 0.2.2-alpha.3 with npm i -g @deepseek-ai/dsh@alpha; latest and next stay on 0.2.1-rc.5. Pin an explicit version if you need reproducible installs.
Official source: https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.2-alpha.3
Related pages: /cli · /cli/install
DeepSeek Harness Creator Mode showcased
On 2 Nov 2026 DeepSeek announced and showcased Creator Mode in DeepSeek Harness (dsh). In Creator Mode you describe a plugin in plain language, select the Creator Mode option, and the agent builds the plugin for use inside Harness. DeepSeek published a prompt collection for the seven demo plugins from its announcement video: 3D gallery of famous landmarks, Transformer 3D visualization, DSH pixelated interface, Poster design workbench, Video editing workbench, Paper reader, and Conversation statistics. DeepSeek notes some demos were refined over several rounds of conversation, so one prompt may not reproduce the video exactly.
What it means: Use the published prompts as starting points; expect iteration if you want results matching the announcement demos.
Official source: https://cdn.deepseek.com/harness/dsh-creator-mode-prompt-collection-en.html
Related pages: /cli · /cli/install
dsh-superagent-prototype open-sourced
DeepSeek's Harness team open-sourced dsh-superagent-prototype under the MIT licence (repository description: 'Towards a More Autonomous Agent System'; first commit 31 Oct 2026 12:05 UTC, message 'feat: add the dsh-superagent research prototype'). Per its README, this is a research prototype for reference only, not an official DeepSeek product. It is a proactive coding agent for one GitHub repository, packaged as a bundle for DeepSeek Harness (dsh). It listens to GitHub and optionally Feishu; when an allowed person asks for work it opens a work item, works in its own agent session and directory, and answers in the original conversation (for example with a pull request). A web panel shows what needs a person. Its stated principle is 'take as little human attention as possible, and give as much value as possible'. It accompanies a DeepSeek Harness article 'Exploring Super-agent Design: Towards a More Autonomous Agent System', also posted by @DeepSeekHarness on X on 1 Nov 2026. The published code differs from DeepSeek's internal version and some designs in the article are not implemented.
What it means: Treat the repository as reference material, not a supported product. Expect gaps versus DeepSeek's internal implementation and versus every design described in the accompanying article.
Official source: https://github.com/deepseek-ai/dsh-superagent-prototype
Related pages: /cli · /cli/install
@deepseek-ai/dsh 0.2.1-rc.5 published on npm
npm dist-tags for @deepseek-ai/dsh list both latest and next as 0.2.1-rc.5, published 29 Oct 2026, while alpha remains 0.2.2-alpha.2 (previously latest pointed to 0.2.1-rc.2). This is a pre-release version, not a stable GA release. GitHub marks dsh-v0.2.1-rc.5 as a pre-release, published 30 Oct 2026. The release notes only improve how thinking content is displayed in conversations.
What it means: npm install -g @deepseek-ai/dsh now installs 0.2.1-rc.5 on latest and next. Install 0.2.2-alpha.2 with npm i -g @deepseek-ai/dsh@alpha. Pin an explicit version if you need reproducible installs.
Official source: https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.1-rc.5
Related pages: /cli · /cli/install
@UnslothAI shared new GGUF quantisations of DeepSeek-V4.1-Flash-1017
@UnslothAI shared new GGUF quantisations of DeepSeek-V4.1-Flash-1017 for running the model locally. Unsloth says they run on machines with 64GB+ RAM and reports that its Dynamic quants score highest on benchmarks such as GPQA Diamond, AIME25 and LiveCodeBench v6.
Shared by @UnslothAI (original post)
Related pages: /models/deepseek-v4-1-flash
@deepseek-ai/dsh 0.2.2-alpha.2 published on npm
npm dist-tags for @deepseek-ai/dsh list alpha as 0.2.2-alpha.2, published 27 Oct 2026, while latest remains 0.2.1-rc.2 and next lists 0.2.1-rc.4. This is an alpha pre-release version, not a stable GA release. GitHub marks dsh-v0.2.2-alpha.2 as a pre-release. The release notes add rich-text plugin details in the plugin manager with support for updating third-party plugin packages and restoring the previous version if hot updates fail; quoting chat text into the composer; copying a complete conversation as Markdown; interface text and locale configuration for Japanese, Spanish, French, German, Russian, and Brazilian Portuguese; confirming to clear all inactive automation tasks at once while keeping active tasks; and experimental plugins including the subagent_oar tool for delegating independent tasks to local Codex or Claude CLIs, Developer tools statistics for per-turn and per-step context, cache, model, and tool measurements, and Change requests and worktrees for running Sessions in separate Git worktrees with linked PR status and code changes. They also include many mobile, narrow-screen, and sidebar layout and navigation improvements, and raise the default per-image upload limit to 50 MiB.
What it means: Install 0.2.2-alpha.2 with npm i -g @deepseek-ai/dsh@alpha; latest stays on 0.2.1-rc.2. Pin an explicit version if you need reproducible installs.
Official source: https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.2-alpha.2
Related pages: /cli · /cli/install
@deepseek-ai/dsh 0.2.1-rc.4 published on npm
npm dist-tags for @deepseek-ai/dsh list next as 0.2.1-rc.4, published 26 Oct 2026, while latest remains 0.2.1-rc.2. This is a pre-release version, not a stable GA release. GitHub marks dsh-v0.2.1-rc.4 as a pre-release. The release notes only fix diagnostics leaking into stdout in ACP mode.
What it means: Install 0.2.1-rc.4 with npm i -g @deepseek-ai/dsh@next; latest stays on 0.2.1-rc.2. Pin an explicit version if you need reproducible installs.
Official source: https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.1-rc.4
Related pages: /cli · /cli/install
@deepseek-ai/dsh 0.2.1-rc.3 published on npm
npm dist-tags for @deepseek-ai/dsh list next as 0.2.1-rc.3, published 23 Oct 2026, while latest remains 0.2.1-rc.2. This is a pre-release version, not a stable GA release. GitHub marks dsh-v0.2.1-rc.3 as a pre-release. The release notes add More menus to the left sidebar, conversation views and a right sidebar for browsing and creating plugins; fix New Session losing an unsent draft after switching sessions; and make the DeepSeek API key adapter update its model list the same way as account sign-in.
What it means: Install 0.2.1-rc.3 with npm i -g @deepseek-ai/dsh@next; latest stays on 0.2.1-rc.2. Pin an explicit version if you need reproducible installs.
Official source: https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.1-rc.3
Related pages: /cli · /cli/install
DeepInfer and deepseek-recipe open-sourced
DeepSeek open-sourced DeepInfer, the in-house inference framework that powers its production services, under the Apache-2.0 licence. Per its README, this release covers offline inference, supports DeepSeek-V4.1-Flash (default checkpoint deepseek-ai/DeepSeek-V4.1-Flash-1017), runs on NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs, and includes optimized kernels (FlashMLA, DeepGEMM, DeepEP, TileKernels, DeepSelect, TileLang), attention DP and context parallelism, MoE expert parallelism with EPLB, speculative decoding (DSpark), SWA bounded replay, asynchronous scheduling, batch-invariant inference, deterministic sampling and structured outputs (regex, JSON schema, EBNF). A KV cache system, prefill/decode disaggregation and load balancing are not included. DeepSeek states that inference code for future models will land in the same repository. On the same date DeepSeek also open-sourced deepseek-recipe under the MIT licence: Rust libraries with Python bindings for production token encoding and decoding, converting Messages, Chat Completions and Responses API requests into DeepSeek V4/V4.1 prompts and token IDs aligned with the official DeepSeek API processing so community deployments match hosted behavior.
What it means: Use the DeepInfer repository for offline inference on supported hardware; pair with deepseek-recipe when you need API-aligned prompts and token IDs. Do not expect KV-cache management, prefill/decode disaggregation or load balancing in this release.
Official source: https://github.com/deepseek-ai/DeepInfer
Related pages: /models/deepseek-v4-1-flash
@PawelHuryn shared follow-up agent-harness bug-fix benchmark for V4.1-Flash-1017
@PawelHuryn shared a follow-up to his own bug-fixing benchmark: he reports running DeepSeek V4.1 Flash 1017 89 times across 9 agent harnesses (high effort, 5 attempts each). His reported scores: Pi 40.4, Grok Build CLI 40.4, DeepSeek Harness 39.2, OpenCode 2.0 38.6, OpenCode 36.6, Claude Code 36.4, Codex 33.2, Hermes Agent 32. In an earlier 18 Oct post he reports that the DeepSeek harness raised the model's score by 2 to 3 points on average but cost about 80% more because of subagents and extra output tokens.
Shared by @PawelHuryn (original post)
Related pages: /cli · /models/deepseek-v4-1-flash
@deepseek-ai/dsh 0.2.1-rc.2 published on npm
npm dist-tags for @deepseek-ai/dsh list both latest and next as 0.2.1-rc.2, published 17 Oct 2026 (0.2.1-rc.1 was published earlier the same day). These are pre-release versions, not a stable GA release. The GitHub release notes for dsh-v0.2.1-rc.2 list support for DeepSeek-V4.1-Flash-1017 and for the DeepSeek Search and Fetch APIs, which can be turned on in Web Search settings. The 2026-10-17 API change log recommends upgrading to DeepSeek Harness 0.2.1-rc.1 or later for the best experience with the new model. Release 0.2.1-rc.1 was the first release candidate in the 0.2.1 line and rolled up changes since 0.2.0-rc.2, including plugin installs from custom plugin index sources, plugin enable, disable, and update handling, and several experimental plugins.
What it means: Treat latest and next as release candidates on the 0.2.1 line; pin an explicit version if you need reproducible installs.
Official source: https://www.npmjs.com/package/@deepseek-ai/dsh
Related pages: /cli · /cli/install
DeepSeek-V4.1-Flash-1017 Update
DeepSeek V4.1 Flash has been upgraded to DeepSeek-V4.1-Flash-1017. The API calling method and pricing remain unchanged — simply use deepseek-flash to access the latest model. DeepSeek also opened its in-house Search / Fetch API on the same date.
What it means: Keep calling deepseek-flash; upgrade DeepSeek Harness to 0.2.1-rc.1 or later if you rely on Harness-optimized behavior per the change log.
Official source: https://api-docs.deepseek.com/updates#date-2026-10-17
Related pages: /guides/model-name-migration · /models/deepseek-v4-1-flash · /news
@Zaddyzaddy shared VulnPR-100 results for V4.1-Flash-1017
@Zaddyzaddy shared that the @BugBunny_ai team gave DeepSeek-V4.1-Flash-1017 the same 100 pull requests with known vulnerabilities in their VulnPR-100 benchmark, with an hour per PR. They report it found 54 for $7.53 and call it the best-performing open-weight model on their leaderboard.
Shared by @Zaddyzaddy (original post)
Related pages: /models/deepseek-v4-1-flash
@PawelHuryn tested V4.1-Flash-1017 on real bug-fixing work
@PawelHuryn shared first results from his own test, two repos with 105 hidden bugs to find and fix. He reports V4.1 Flash 1017 (max) scored 34, against 41.7 for Opus 5.5 (max), 40.7 for Gemini 4 Argon (high) and 29 for Grok 4.7 (xhigh) in his run.
Shared by @PawelHuryn (original post)
Related pages: /models/deepseek-v4-1-flash
@Bhavani_00007 compared cost on a rocket launch scene
@Bhavani_00007 shared a side-by-side where V4.1 Flash 1017 and GPT 6.1 Sol each built a rocket launch scene. He reports the DeepSeek run cost $0.78 and took 3 minutes, against $6.10 and 5 minutes for GPT 6.1 Sol.
Shared by @Bhavani_00007 (original post)
Related pages: /api-pricing · /models/deepseek-v4-1-flash
DeepSeek Harness desktop apps for macOS and Windows
DeepSeek lists packaged DeepSeek Harness desktop installers for macOS (Apple Silicon) and Windows on its official download and Harness product pages. Direct downloads are published at download.deepseek.com (macOS .dmg and Windows .exe). The dsh-v0.2.0-rc.2 GitHub release documents macOS and Windows desktop features, including a bundled dsh command exposed from the menu bar. Linux is not offered as a packaged desktop build in these official channels; use the @deepseek-ai/dsh npm package instead.
What it means: Use the official download page for current desktop installers; treat this as part of the 0.2.0-rc preview line documented on GitHub, not a separate stable GA desktop channel.
Official source: https://www.deepseek.com/en/download/
@deepseek-ai/dsh 0.2.0-rc.2 published on npm
npm dist-tags for @deepseek-ai/dsh list latest as 0.2.0-rc.2, next as 0.2.1-rc.2, and alpha as 0.2.1-alpha.5. These are pre-release versions, not a stable major/minor GA release.
What it means: Treat latest as a release candidate; pin an explicit version if you need reproducible installs.
Official source: https://www.npmjs.com/package/@deepseek-ai/dsh
Related pages: /cli · /cli/install
DeepSeek-V4.1-Flash Release
DeepSeek's change log says: Today, we officially release the DeepSeek-V4.1-Flash model. It is the smallest model in our new architecture family, with native multimodal visual understanding. The new architecture is designed for a higher capability ceiling, faster inference, higher throughput, and scaling to larger models. API: use model name deepseek-flash for V4.1 Flash; deepseek-v4-flash and deepseek-v4-flash-vision-exp are retired and temporarily routed to V4.1 Flash. V4 Pro API service continues after 2026-09-14 with unchanged billing; API prices were reduced.
What it means: Point integrations at deepseek-flash; re-check costs on the official pricing page; legacy model names still work via temporary routing.
Official source: https://api-docs.deepseek.com/updates#date-2026-09-10
Related pages: /models/deepseek-v4-1-flash · /api-pricing · /guides/model-name-migration · /news
DeepSeek-V4-Flash-Vision-Exp Release
DeepSeek's change log says: Today, the new multimodal vision understanding model DeepSeek-V4-Flash-Vision-Exp is now available on the DeepSeek API platform. This is an experimental model that can be accessed by setting model='deepseek-v4-flash-vision-exp'.
Official source: https://api-docs.deepseek.com/updates#date-2026-08-21
Related pages: /guides/model-name-migration · /models/deepseek-v4-1-flash
DeepSeek-V4-Pro Update
DeepSeek's change log says: The GA release of DeepSeek-V4-Pro has been rolled out on the APP, Web, and API. The API calling method remains unchanged — simply set the model name to deepseek-v4-pro to use the latest version.
Official source: https://api-docs.deepseek.com/updates#date-2026-08-13
Related pages: /guides/model-name-migration · /models/deepseek-v4-pro
DeepSeek-V4-Flash Update
DeepSeek's change log says: The official release of the DeepSeek-V4-Flash API is now in public beta. The API calling method remains unchanged — simply set the model name to deepseek-v4-flash to use the latest version.
Official source: https://api-docs.deepseek.com/updates#date-2026-07-31
Related pages: /guides/model-name-migration · /models/deepseek-v4-1-flash
DeepSeek-V4
DeepSeek's change log says: The DeepSeek API now supports V4-Pro and V4-Flash, available via both the OpenAI ChatCompletions interface and the Anthropic interface. To access the new models, the base_url remains unchanged, and the model parameter should be set to deepseek-v4-pro or deepseek-v4-flash.
Official source: https://api-docs.deepseek.com/updates#date-2026-04-24
Related pages: /models · /api-pricing · /guides/model-name-migration · /guides/off-peak-pricing · /news
2025
DeepSeek-V3.2-Speciale
DeepSeek-V3.2-Speciale is served via a temporary endpoint: base_url="https://api.deepseek.com/v3.2_speciale_expires_on_20251215". Same pricing as V3.2, no tool calls, available until Dec 15th, 2025, 15:59 (UTC Time).
Official source: https://api-docs.deepseek.com/updates#date-2025-12-01
Related pages: /guides/model-name-migration · /models
DeepSeek-V3.2
DeepSeek's change log says: Both deepseek-chat and deepseek-reasoner have been upgraded to DeepSeek-V3.2.
Official source: https://api-docs.deepseek.com/updates#date-2025-12-01
Related pages: /guides/model-name-migration · /models · /news
DeepSeek-V3.2-Exp
DeepSeek's change log says: Both deepseek-chat and deepseek-reasoner have been upgraded to DeepSeek-V3.2-Exp.
Official source: https://api-docs.deepseek.com/updates#date-2025-09-29
Related pages: /guides/model-name-migration · /models
DeepSeek-V3.1-Terminus
DeepSeek's change log says: Both deepseek-chat and deepseek-reasoner have been upgraded to DeepSeek-V3.1-Terminus. deepseek-chat corresponds to DeepSeek-V3.1-Terminus's non-thinking mode, while deepseek-reasoner corresponds to its thinking mode.
Official source: https://api-docs.deepseek.com/updates#date-2025-09-22
Related pages: /guides/model-name-migration · /models
DeepSeek-V3.1
DeepSeek's change log says: Both deepseek-chat and deepseek-reasoner have been upgraded to DeepSeek-V3.1. deepseek-chat corresponds to DeepSeek-V3.1's non-thinking mode, while deepseek-reasoner corresponds to its thinking mode.
Official source: https://api-docs.deepseek.com/updates#date-2025-08-21
Related pages: /guides/model-name-migration · /models
deepseek-reasoner
deepseek-reasoner Model Upgraded to DeepSeek-R1-0528:
Official source: https://api-docs.deepseek.com/updates#date-2025-05-28
Related pages: /guides/model-name-migration · /models/deepseek-r1
deepseek-chat
deepseek-chat Model Upgraded to DeepSeek-V3-0324:
Official source: https://api-docs.deepseek.com/updates#date-2025-03-24
Related pages: /guides/model-name-migration · /models/deepseek-v3
deepseek-reasoner
DeepSeek's change log says: deepseek-reasoner is our new model DeepSeek-R1. You can invoke DeepSeek-R1 by specifying model='deepseek-reasoner'. For details, please refer to: DeepSeek-R1 Release
What it means: deepseek-reasoner now serves DeepSeek-R1; see official R1 release notes.
Official source: https://api-docs.deepseek.com/updates#date-2025-01-20
Related pages: /guides/model-name-migration · /models/deepseek-r1 · /news
2024
deepseek-chat
DeepSeek's change log says: The deepseek-chat model has been upgraded to DeepSeek-V3. The API remains unchanged. You can invoke DeepSeek-V3 by specifying model='deepseek-chat'. For details, please refer to: introducing DeepSeek-V3
What it means: deepseek-chat now serves DeepSeek-V3 with the same API surface.
Official source: https://api-docs.deepseek.com/updates#date-2024-12-26
Related pages: /guides/model-name-migration · /models/deepseek-v3 · /news
deepseek-chat
DeepSeek's change log says: The deepseek-chat model has been upgraded to DeepSeek-V2.5-1210, with improvements across various capabilities. See the official change log for benchmark tables.
Official source: https://api-docs.deepseek.com/updates#date-2024-12-10
Related pages: /guides/model-name-migration · /models
deepseek-coder & deepseek-chat upgraded to DeepSeek V2.5
The DeepSeek V2 Chat and DeepSeek Coder V2 models were merged and upgraded into DeepSeek V2.5. DeepSeek states the API remains backward compatible for existing integrations.
What it means: No API parameter change required per the change log; verify behavior if you relied on separate chat vs coder endpoints.
Official source: https://api-docs.deepseek.com/updates#date-2024-09-05
Related pages: /guides/model-name-migration · /models
API Launches Context Caching on Disk Technology
DeepSeek's change log says: The DeepSeek API has innovatively adopted hard disk caching, reducing prices by another order of magnitude.
What it means: Review official pricing for cache-hit rates after disk context caching launched.
Official source: https://api-docs.deepseek.com/updates#date-2024-08-02
Related pages: /api-pricing · /guides/off-peak-pricing · /news
New API Features
For more details, please check the documentation New API Features 2024/07/25
Official source: https://api-docs.deepseek.com/updates#date-2024-07-25
Related pages: /models
deepseek-coder
The deepseek-coder model has been upgraded to DeepSeek-Coder-V2-0724.
Official source: https://api-docs.deepseek.com/updates#date-2024-07-24
Related pages: /guides/model-name-migration · /models
deepseek-chat
DeepSeek's change log says: The deepseek-chat model has been upgraded to DeepSeek-V2-0628.
Official source: https://api-docs.deepseek.com/updates#date-2024-06-28
Related pages: /guides/model-name-migration · /models
deepseek-coder
The deepseek-coder model has been upgraded to DeepSeek-Coder-V2-0614, significantly enhancing its coding capabilities. It has reached the level of GPT-4-Turbo-0409 in code generation, code understanding, code debugging, and code completion. Additionally, it possesses excellent mathematical and reasoning abilities, and its general capabilities are on par with DeepSeek-V2-0517.
Official source: https://api-docs.deepseek.com/updates#date-2024-06-14
Related pages: /guides/model-name-migration · /models
deepseek-chat
DeepSeek's change log says: The deepseek-chat model has been upgraded to DeepSeek-V2-0517. The model has seen a significant improvement in following instructions, with the IFEval Benchmark Prompt-Level accuracy jumping from 63.9% to 77.6%. Additionally, on API end, we have optimized model ability to follow instruction filled in the ``system" part. This optimization has significantly elevated the user experience across a variety of tasks, including immersive translation, Retrieval-Augmented Generation (RAG), and more.
Official source: https://api-docs.deepseek.com/updates#date-2024-05-17
Related pages: /guides/model-name-migration · /models
API and pricing change history
See also the live API pricing directory, off-peak hours guide, and the official pricing page.
DeepInfer and deepseek-recipe open-sourced
DeepSeek open-sourced DeepInfer, the in-house inference framework that powers its production services, under the Apache-2.0 licence. Per its README, this release covers offline inference, supports DeepSeek-V4.1-Flash (default checkpoint deepseek-ai/DeepSeek-V4.1-Flash-1017), runs on NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs, and includes optimized kernels (FlashMLA, DeepGEMM, DeepEP, TileKernels, DeepSelect, TileLang), attention DP and context parallelism, MoE expert parallelism with EPLB, speculative decoding (DSpark), SWA bounded replay, asynchronous scheduling, batch-invariant inference, deterministic sampling and structured outputs (regex, JSON schema, EBNF). A KV cache system, prefill/decode disaggregation and load balancing are not included. DeepSeek states that inference code for future models will land in the same repository. On the same date DeepSeek also open-sourced deepseek-recipe under the MIT licence: Rust libraries with Python bindings for production token encoding and decoding, converting Messages, Chat Completions and Responses API requests into DeepSeek V4/V4.1 prompts and token IDs aligned with the official DeepSeek API processing so community deployments match hosted behavior.
Official source: https://github.com/deepseek-ai/DeepInfer
Related pages: /models/deepseek-v4-1-flash
API Launches Context Caching on Disk Technology
DeepSeek's change log says: The DeepSeek API has innovatively adopted hard disk caching, reducing prices by another order of magnitude.
Official source: https://api-docs.deepseek.com/updates#date-2024-08-02
Related pages: /api-pricing · /guides/off-peak-pricing · /news
New API Features
For more details, please check the documentation New API Features 2024/07/25
Official source: https://api-docs.deepseek.com/updates#date-2024-07-25
Related pages: /models
DeepSeek Harness (dsh) releases
Stable vs pre-release: npm latest may point at an rc build. See CLI directory and install guide.
@deepseek-ai/dsh 0.2.2-alpha.3 published on npm
npm dist-tags for @deepseek-ai/dsh list alpha as 0.2.2-alpha.3, published 3 Nov 2026 07:48 UTC, while latest and next remain 0.2.1-rc.5. This is an alpha pre-release version, not a stable GA release. GitHub marks dsh-v0.2.2-alpha.3 as a pre-release. The release notes unify mouse and keyboard context menus across Sessions, workspaces, and the sidebar; keep a directory tree in file previews; unify code review and worktree access in the experimental change-request and worktree feature (searchable file tree and per-file navigation for local and PR diffs, quote selected code and comments to the Agent, track checks and reviews), folding the standalone Git worktree plugin into it (users who enabled @deepseek-ai/dsh-experimental-tool-worktree must remove it from dsh.profile.bundles); move automation tasks to unified host-level scheduled-task management with manual creation, direct editing, and session-based creation; add a new apply_patch tool (off by default); extend TypeScript and Python SDKs with native commands and plugin management; turn read-before-write and external-change checks off by default while keeping sandbox limits and Review; and include several bug fixes.
Official source: https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.2-alpha.3
Related pages: /cli · /cli/install
DeepSeek Harness Creator Mode showcased
On 2 Nov 2026 DeepSeek announced and showcased Creator Mode in DeepSeek Harness (dsh). In Creator Mode you describe a plugin in plain language, select the Creator Mode option, and the agent builds the plugin for use inside Harness. DeepSeek published a prompt collection for the seven demo plugins from its announcement video: 3D gallery of famous landmarks, Transformer 3D visualization, DSH pixelated interface, Poster design workbench, Video editing workbench, Paper reader, and Conversation statistics. DeepSeek notes some demos were refined over several rounds of conversation, so one prompt may not reproduce the video exactly.
Official source: https://cdn.deepseek.com/harness/dsh-creator-mode-prompt-collection-en.html
Related pages: /cli · /cli/install
dsh-superagent-prototype open-sourced
DeepSeek's Harness team open-sourced dsh-superagent-prototype under the MIT licence (repository description: 'Towards a More Autonomous Agent System'; first commit 31 Oct 2026 12:05 UTC, message 'feat: add the dsh-superagent research prototype'). Per its README, this is a research prototype for reference only, not an official DeepSeek product. It is a proactive coding agent for one GitHub repository, packaged as a bundle for DeepSeek Harness (dsh). It listens to GitHub and optionally Feishu; when an allowed person asks for work it opens a work item, works in its own agent session and directory, and answers in the original conversation (for example with a pull request). A web panel shows what needs a person. Its stated principle is 'take as little human attention as possible, and give as much value as possible'. It accompanies a DeepSeek Harness article 'Exploring Super-agent Design: Towards a More Autonomous Agent System', also posted by @DeepSeekHarness on X on 1 Nov 2026. The published code differs from DeepSeek's internal version and some designs in the article are not implemented.
Official source: https://github.com/deepseek-ai/dsh-superagent-prototype
Related pages: /cli · /cli/install
@deepseek-ai/dsh 0.2.1-rc.5 published on npm
npm dist-tags for @deepseek-ai/dsh list both latest and next as 0.2.1-rc.5, published 29 Oct 2026, while alpha remains 0.2.2-alpha.2 (previously latest pointed to 0.2.1-rc.2). This is a pre-release version, not a stable GA release. GitHub marks dsh-v0.2.1-rc.5 as a pre-release, published 30 Oct 2026. The release notes only improve how thinking content is displayed in conversations.
Official source: https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.1-rc.5
Related pages: /cli · /cli/install
@deepseek-ai/dsh 0.2.2-alpha.2 published on npm
npm dist-tags for @deepseek-ai/dsh list alpha as 0.2.2-alpha.2, published 27 Oct 2026, while latest remains 0.2.1-rc.2 and next lists 0.2.1-rc.4. This is an alpha pre-release version, not a stable GA release. GitHub marks dsh-v0.2.2-alpha.2 as a pre-release. The release notes add rich-text plugin details in the plugin manager with support for updating third-party plugin packages and restoring the previous version if hot updates fail; quoting chat text into the composer; copying a complete conversation as Markdown; interface text and locale configuration for Japanese, Spanish, French, German, Russian, and Brazilian Portuguese; confirming to clear all inactive automation tasks at once while keeping active tasks; and experimental plugins including the subagent_oar tool for delegating independent tasks to local Codex or Claude CLIs, Developer tools statistics for per-turn and per-step context, cache, model, and tool measurements, and Change requests and worktrees for running Sessions in separate Git worktrees with linked PR status and code changes. They also include many mobile, narrow-screen, and sidebar layout and navigation improvements, and raise the default per-image upload limit to 50 MiB.
Official source: https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.2-alpha.2
Related pages: /cli · /cli/install
@deepseek-ai/dsh 0.2.1-rc.4 published on npm
npm dist-tags for @deepseek-ai/dsh list next as 0.2.1-rc.4, published 26 Oct 2026, while latest remains 0.2.1-rc.2. This is a pre-release version, not a stable GA release. GitHub marks dsh-v0.2.1-rc.4 as a pre-release. The release notes only fix diagnostics leaking into stdout in ACP mode.
Official source: https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.1-rc.4
Related pages: /cli · /cli/install
@deepseek-ai/dsh 0.2.1-rc.3 published on npm
npm dist-tags for @deepseek-ai/dsh list next as 0.2.1-rc.3, published 23 Oct 2026, while latest remains 0.2.1-rc.2. This is a pre-release version, not a stable GA release. GitHub marks dsh-v0.2.1-rc.3 as a pre-release. The release notes add More menus to the left sidebar, conversation views and a right sidebar for browsing and creating plugins; fix New Session losing an unsent draft after switching sessions; and make the DeepSeek API key adapter update its model list the same way as account sign-in.
Official source: https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.1-rc.3
Related pages: /cli · /cli/install
@deepseek-ai/dsh 0.2.1-rc.2 published on npm
npm dist-tags for @deepseek-ai/dsh list both latest and next as 0.2.1-rc.2, published 17 Oct 2026 (0.2.1-rc.1 was published earlier the same day). These are pre-release versions, not a stable GA release. The GitHub release notes for dsh-v0.2.1-rc.2 list support for DeepSeek-V4.1-Flash-1017 and for the DeepSeek Search and Fetch APIs, which can be turned on in Web Search settings. The 2026-10-17 API change log recommends upgrading to DeepSeek Harness 0.2.1-rc.1 or later for the best experience with the new model. Release 0.2.1-rc.1 was the first release candidate in the 0.2.1 line and rolled up changes since 0.2.0-rc.2, including plugin installs from custom plugin index sources, plugin enable, disable, and update handling, and several experimental plugins.
Official source: https://www.npmjs.com/package/@deepseek-ai/dsh
Related pages: /cli · /cli/install
DeepSeek Harness desktop apps for macOS and Windows
DeepSeek lists packaged DeepSeek Harness desktop installers for macOS (Apple Silicon) and Windows on its official download and Harness product pages. Direct downloads are published at download.deepseek.com (macOS .dmg and Windows .exe). The dsh-v0.2.0-rc.2 GitHub release documents macOS and Windows desktop features, including a bundled dsh command exposed from the menu bar. Linux is not offered as a packaged desktop build in these official channels; use the @deepseek-ai/dsh npm package instead.
Official source: https://www.deepseek.com/en/download/
@deepseek-ai/dsh 0.2.0-rc.2 published on npm
npm dist-tags for @deepseek-ai/dsh list latest as 0.2.0-rc.2, next as 0.2.1-rc.2, and alpha as 0.2.1-alpha.5. These are pre-release versions, not a stable major/minor GA release.
Official source: https://www.npmjs.com/package/@deepseek-ai/dsh
Related pages: /cli · /cli/install
Papers and tech reports
When DeepSeek publishes a tech report with substance, it will appear here with an arXiv or official doc link. None are listed in this timeline yet.
How we verify
Nothing is added until an official DeepSeek source confirms it (API change log, API news posts, pricing page, github.com/deepseek-ai, or npm @deepseek-ai/dsh). Social posts are alerts only — they are not published as facts here.
FAQ
- How often is this page updated?
- When the official API change log or other verified sources change; entries include a checked date.
- Where is the RSS feed?
- DeepSeek news RSS feed
- Is V5 announced?
- Not in any official source we track; we do not speculate on future models.