Note: This article is not an official DeepSeek roadmap; it is a source-checked status tracker based on public releases and reputable reporting.
As of May 7, 2026, the safest way to understand the DeepSeek Roadmap is not to look for a traditional public product roadmap. DeepSeek does not currently publish a single forward-looking roadmap page with guaranteed future release dates. Instead, its roadmap is visible through official release notes, API documentation, model pricing pages, change logs, model cards, Hugging Face repositories, GitHub materials, and reputable reporting.
The clearest confirmed milestone is DeepSeek-V4 Preview, officially announced on April 24, 2026. DeepSeek says V4 Preview is live, open-sourced, available on chat, app, and API, and split into DeepSeek-V4-Pro and DeepSeek-V4-Flash. The company also says both V4 API models support a 1M-token context and Thinking / Non-Thinking modes.
The most important caution: DeepSeek itself reminds readers to rely only on official accounts for DeepSeek news, and says statements from other channels do not reflect its views. That warning matters because DeepSeek Roadmap Rumors around R2, V5, multimodal features, and hardware partnerships often spread before official confirmation.
Quick Answer
What is official now?
DeepSeek-V4 Preview is official. DeepSeek-V4-Pro and DeepSeek-V4-Flash are listed in official docs, API documentation, pricing, and model listings. The API supports deepseek-v4-pro and deepseek-v4-flash.
What is still rumor?
A confirmed DeepSeek R2 release date, a DeepSeek V5 roadmap, and a final V4 GA timeline are not officially confirmed in DeepSeek’s current public docs checked for this article. Reuters reported in June 2025 that R2 timing had not been determined, citing The Information.
What should readers watch next?
Watch the official DeepSeek API Docs News page, Change Log, Models & Pricing page, and verified Hugging Face organization for releases, model IDs, pricing, context limits, open weights, and deprecation notices.
What should developers do now?
Test V4-Pro and V4-Flash separately, migrate away from old aliases, check current pricing before production use, and keep rollback options because V4 is still labeled “Preview.”
Table of Contents
- DeepSeek Roadmap at a Glance
- What DeepSeek Has Officially Released
- DeepSeek V4 Preview: What Is Official
- DeepSeek Roadmap Rumors: What Is Not Official
- Official vs Rumor: How to Verify Any DeepSeek Roadmap Claim
- What Developers Should Do Now
- What Businesses and AI Watchers Should Watch Next
- DeepSeek Roadmap FAQ
- Image Suggestions
- Sources Checked
- JSON-LD Schema
DeepSeek Roadmap at a Glance
The table below separates official releases from reported claims and unsupported speculation.
| Status | Item | What we know | Source type | Confidence |
|---|---|---|---|---|
| Official | DeepSeek-V4 Preview | Officially released on April 24, 2026; live on chat, app, and API; open weights available. | DeepSeek API Docs | Confirmed |
| Official | DeepSeek-V4-Pro | DeepSeek describes it as the larger V4 variant; official release notes list 1.6T total parameters and 49B active parameters. | DeepSeek API Docs / model card | Confirmed |
| Official | DeepSeek-V4-Flash | DeepSeek describes it as the faster, more economical V4 variant; release notes list 284B total and 13B active parameters. | DeepSeek API Docs / model card | Confirmed |
| Official | 1M context support | Official pricing and model details list a 1M context length for V4 models. | DeepSeek Models & Pricing | Confirmed |
| Official | API model IDs | Current model IDs include deepseek-v4-flash and deepseek-v4-pro. | DeepSeek API reference | Confirmed |
| Official | Open weights / model cards | DeepSeek links V4 open weights, and the verified Hugging Face organization lists V4-Pro, V4-Flash, and base variants. | DeepSeek / Hugging Face | Confirmed |
| Official | Legacy model alias deprecation | deepseek-chat and deepseek-reasoner are scheduled for deprecation on July 24, 2026. | DeepSeek Change Log / API docs | Confirmed |
| Reported | DeepSeek R2 | Reuters reported that R2 was planned as an R1 successor but later reported that timing was not determined. | Reuters / The Information report | Reported |
| Unsupported | DeepSeek V5 or future versions | No official DeepSeek V5 announcement was found in the checked official release pages. | Absence from official docs checked | Unsupported |
| Rumor | Native V4 multimodal claims | DeepSeek previously hinted at future multimodal support after V3, but current V4 docs checked here do not confirm native V4 image/audio/video support. | Official hint vs current docs | Rumor |
| Reported | Huawei hardware adaptation / support | Reuters reported V4 was adapted for Huawei chips, and Huawei said its Ascend supernode would support V4. This is not the same as a DeepSeek-published roadmap. | Reuters | Reported |
DeepSeek’s V4 release page supports the official rows above: it names V4-Pro and V4-Flash, says V4 Preview is live and open-sourced, lists 1M context, and says the API is available. The API model list confirms deepseek-v4-flash and deepseek-v4-pro as current model identifiers.
What DeepSeek Has Officially Released
DeepSeek’s official release history shows a pattern: the company tends to announce what has shipped, rather than publish long-dated promises. The current official news navigation lists V4 Preview as the latest release, followed by V3.2, V3.2-Exp, V3.1-Terminus, V3.1, R1 updates, the app, V3, R1-Lite, and V2.5 releases.
| Date | Release | Official status | Why it matters |
|---|---|---|---|
| Sep. 5, 2024 | DeepSeek-V2.5 | Official launch | Combined DeepSeek-V2-0628 and DeepSeek-Coder-V2-0724, available on web/API, and open-sourced on Hugging Face. |
| Nov. 20, 2024 | DeepSeek-R1-Lite-Preview | Official preview | Early reasoning-focused preview with real-time thought process shown in the interface; open-source models and API were described as coming soon. |
| Dec. 10, 2024 | DeepSeek-V2.5-1210 | Official update | Added web internet search and marked the end of the V2.5 series. |
| Dec. 26, 2024 | DeepSeek-V3 | Official release | Introduced a 671B-parameter MoE model with 37B active parameters and open-source model/paper links. |
| Jan. 15, 2025 | DeepSeek App | Official app release | Officially available through App Store, Google Play, and major Android markets, with chat sync, web search, DeepThink mode, and file upload. |
| Jan. 20, 2025 | DeepSeek-R1 | Official release | Open-source reasoning model under MIT license, with API access through deepseek-reasoner. |
| Mar. 25, 2025 | DeepSeek-V3-0324 | Official update | Improved reasoning, front-end development, and tool-use capabilities; released under MIT license with open weights. |
| May 28, 2025 | DeepSeek-R1-0528 | Official update | Improved benchmark performance, reduced hallucinations, and added JSON output and function calling support. |
| Aug. 21, 2025 | DeepSeek-V3.1 | Official release | Introduced hybrid Think / Non-Think inference, 128K context, agent upgrades, and Anthropic API format support. |
| Sep. 22, 2025 | DeepSeek-V3.1-Terminus | Official update | Addressed user feedback, language consistency, and agent performance; available on app, web, and API. |
| Sep. 29, 2025 | DeepSeek-V3.2-Exp | Official experimental release | Debuted DeepSeek Sparse Attention for long-context efficiency and was live on app, web, and API. |
| Dec. 1, 2025 | DeepSeek-V3.2 | Official release | Reasoning-first model built for agents; V3.2 became the official successor to V3.2-Exp. |
| Apr. 24, 2026 | DeepSeek-V4 Preview | Official preview release | Introduced V4-Pro and V4-Flash, 1M default context, open weights, and API availability. |
No newer official model release than DeepSeek-V4 Preview was found in the checked DeepSeek API Docs News list as of May 7, 2026.
DeepSeek V4 Preview: What Is Official
DeepSeek-V4 Preview is the central confirmed item in the DeepSeek roadmap 2026 conversation. DeepSeek’s release note says V4 Preview is officially live and open-sourced, with V4-Pro and V4-Flash available through chat, app, and API.
The two official V4 variants are:
DeepSeek-V4-Pro
This is the higher-capability V4 variant. DeepSeek’s release page lists it as 1.6T total parameters with 49B active parameters. The V4 model card also lists Pro as 1.6T total with 49B activated for each token.
DeepSeek-V4-Flash
This is the faster and more economical V4 variant. DeepSeek’s release page lists Flash as 284B total parameters with 13B active parameters. The V4 model card lists Flash at 285B parameters with 13B activated for each token, so publishers should avoid overinterpreting the one-billion difference and simply cite the official document used.
The API side is also confirmed. DeepSeek’s Change Log says the API now supports V4-Pro and V4-Flash through both the OpenAI ChatCompletions interface and the Anthropic interface. The base URL remains unchanged, and developers should set the model parameter to deepseek-v4-pro or deepseek-v4-flash.
DeepSeek’s pricing page lists both V4 models with 1M context length, 384K maximum output, JSON output, tool calls, chat prefix completion, and FIM completion in non-thinking mode only. It also says both models support non-thinking and thinking modes, with thinking mode as the default.
Current official pricing, as checked on May 7, 2026, is listed per 1M tokens. V4-Flash is shown at $0.0028 per 1M input tokens for cache hits, $0.14 for cache misses, and $0.28 for output. V4-Pro is shown at discounted prices of $0.003625 for cache-hit input, $0.435 for cache-miss input, and $0.87 for output, with the page noting a 75% V4-Pro discount extended until May 31, 2026. DeepSeek also warns that product prices may vary and recommends checking the pricing page regularly.
Open weights are official. DeepSeek links V4 open weights from its V4 release page, and the verified Hugging Face DeepSeek organization lists the DeepSeek-V4 collection with V4-Flash, V4-Pro, and base variants. The V4 model card says V4 is distributed through open-source repositories and API, and that assets distributed through open-source repositories, including model weights and code, are under the MIT License.
However, Preview still matters. The official title is “DeepSeek V4 Preview Release,” not “DeepSeek V4 final GA.” DeepSeek gives a clear release date and API migration guidance, but it does not provide a final general-availability date in the checked V4 release note.
Developers also need to plan for alias retirement. DeepSeek says deepseek-chat and deepseek-reasoner will be fully retired and inaccessible after July 24, 2026, 15:59 UTC. During the transition, these names route to V4-Flash non-thinking and thinking modes respectively.
DeepSeek Roadmap Rumors: What Is Not Official
This section ranks the most common DeepSeek Roadmap Rumors by confidence. The guiding rule is simple: if DeepSeek has not announced it through an official release note, API doc, model card, verified repository, or official account, it should not be treated as confirmed.
DeepSeek R2 Release Date Rumors
Status: Reported, not officially released.
DeepSeek R2 has been one of the most discussed roadmap topics since R1. Reuters reported in February 2025 that DeepSeek was accelerating the launch of an R2 successor to R1, with sources saying the company had planned an early May release and wanted it out as soon as possible. Reuters also reported that the company hoped R2 would improve coding and reasoning in languages beyond English.
But the story changed. In June 2025, Reuters reported that DeepSeek had not determined R2 timing because CEO Liang Wenfeng was not satisfied with its performance, citing The Information. That same report said engineers were refining R2 until Liang approved release.
As of May 7, 2026, R2 is not listed among the current model IDs in the checked official DeepSeek API model list, which shows deepseek-v4-flash and deepseek-v4-pro. Therefore, any “DeepSeek R2 release date” article should say reported, expected, or unconfirmed, unless a new official DeepSeek source says otherwise.
V4 Final Release or GA Timeline
Status: Unconfirmed.
DeepSeek has confirmed V4 Preview, not a final V4 GA timeline. The official release note calls it “DeepSeek V4 Preview Release,” and the checked API docs provide model IDs, pricing, context length, and deprecation instructions, but not a final GA date.
For production users, that means V4 can be used now, but teams should monitor the Change Log and keep migration plans flexible.
Multimodal DeepSeek Claims
Status: Rumor for V4 native multimodal, unless separately confirmed.
DeepSeek did hint at future multimodal support when announcing V3 in December 2024, saying users could look forward to multimodal support and other cutting-edge features in the DeepSeek ecosystem.
That statement is not the same as a current confirmation that DeepSeek V4 Preview is a native multimodal model. The checked V4 API pricing and model details list text-oriented API capabilities such as JSON output, tool calls, chat prefix completion, and FIM completion, but do not confirm native V4 image, audio, or video input/output.
DeepSeek does have separate multimodal or vision-related projects on its verified Hugging Face organization, including DeepSeek-OCR and Janus collections/spaces, but those should not be confused with confirmed native V4 multimodal support.
DeepSeek V5 or Future Roadmap Claims
Status: Unsupported.
No official DeepSeek V5 announcement was found in the checked DeepSeek API Docs News list, Change Log, pricing page, or current model list. The correct wording is: DeepSeek V5 is not confirmed by the official sources checked for this article.
Hardware and Huawei-Related Claims
Status: Reported, not purely official DeepSeek roadmap.
Reuters reported that DeepSeek-V4 was adapted to run on Huawei chips, and another Reuters item reported that Huawei said its Ascend supernode based on Ascend 950 AI chips would fully support DeepSeek V4 versions after the preview launch.
That is meaningful reporting for infrastructure watchers. But for an article about the DeepSeek V4 roadmap, it should be labeled as reported hardware support/adaptation, not as a DeepSeek-published product roadmap unless DeepSeek itself makes the same claim in official documentation.
Official vs Rumor: How to Verify Any DeepSeek Roadmap Claim
A reliable DeepSeek roadmap article should use a source hierarchy. DeepSeek’s own V4 release note explicitly says readers should rely only on official accounts for DeepSeek news.
| Reliability level | Source | Best use | How to treat it |
|---|---|---|---|
| Highest | Official DeepSeek API Docs News | Release dates, model launches, migration notes | Confirmed if clearly stated |
| Highest | DeepSeek Change Log | API updates, alias changes, model migrations | Confirmed for API behavior |
| Highest | DeepSeek Models & Pricing | Current model IDs, context length, pricing, features | Confirmed but check frequently |
| High | Verified Hugging Face deepseek-ai organization | Open weights, model cards, repository availability | Confirmed for repository presence |
| High | GitHub deepseek-ai repositories | Technical reports, code, model papers | Confirmed for repository content |
| Medium | Reuters, AP, TechCrunch, The Information via Reuters | Reported business, launch, and hardware developments | Reported, not official |
| Low | Reddit, X, Discord, forums, anonymous leaks | Community sentiment and speculation | Rumor unless independently verified |
| Lowest | SEO blogs without primary sourcing | Keyword-driven speculation | Do not use for factual claims |
A simple verification workflow:
First, check the DeepSeek API Docs News page and Change Log. If a model is not there, check the Models & Pricing page and the /models API documentation. Then verify open-weight claims on the verified DeepSeek Hugging Face organization. If the claim is about company strategy, fundraising, hardware, or delayed launches, use reputable reporting and label it as reported.
What Developers Should Do Now
Developers should treat V4-Flash and V4-Pro as separate deployment choices, not interchangeable names. V4-Flash is positioned as the faster and cheaper option, while V4-Pro is the larger, higher-capability option. The right choice depends on task complexity, latency needs, and output-token cost.
Do not rely on deepseek-chat or deepseek-reasoner long term. DeepSeek’s docs say these names will be deprecated on July 24, 2026, and currently correspond to V4-Flash non-thinking and thinking modes.
Benchmark on your own workloads. Official benchmark language is useful for orientation, but production behavior depends on your prompts, tools, files, retrieval system, latency limits, context length, and failure tolerance.
Check pricing before production deployment. DeepSeek’s pricing page says product prices may vary and recommends regularly checking the pricing page for the most recent pricing information.
Keep a rollback path. Because the current V4 release is labeled Preview, teams should track API behavior, prompt compatibility, tool-call behavior, and reasoning-mode costs before committing critical workflows.
What Businesses and AI Watchers Should Watch Next
The biggest item to watch is whether DeepSeek officially releases R2 or whether R2 is effectively superseded by V4-era reasoning modes. At the moment, R2 remains a reported but not officially confirmed release in the sources checked here.
Businesses should also watch for a V4 preview-to-final transition, alias deprecations, pricing changes, third-party benchmarks, and independent evaluations. For regulated environments, safety, privacy, and data-handling reviews matter because Reuters has noted that DeepSeek has faced scrutiny in some countries over security and privacy practices.
Infrastructure watchers should track hardware support and platform partnerships, especially around Huawei Ascend compatibility. But they should distinguish between DeepSeek official product claims and third-party or partner-reported infrastructure claims.
Open-weight users should monitor licensing and model-card updates. The V4 model card says open-source repository assets are MIT licensed, while API access is governed by DeepSeek Open Platform terms.
DeepSeek Roadmap FAQ
Does DeepSeek have an official roadmap?
DeepSeek does not appear to publish a single traditional public roadmap page. Its practical roadmap is visible through official release notes, API docs, change logs, pricing pages, model cards, and verified repositories.
Has DeepSeek V4 been released?
Yes. DeepSeek officially released DeepSeek-V4 Preview on April 24, 2026, with V4-Pro and V4-Flash.
Is DeepSeek V4 Preview the final version?
Not based on the checked official wording. DeepSeek calls it V4 Preview, and no final GA timeline was found in the checked V4 release page or pricing docs.
What is the difference between DeepSeek-V4-Pro and DeepSeek-V4-Flash?
V4-Pro is the larger, higher-capability model. V4-Flash is the faster, more economical option. DeepSeek positions Flash as cost-effective and Pro as stronger for demanding tasks.
Is DeepSeek R2 officially released?
No official R2 release was found in the checked DeepSeek model list or release pages as of May 7, 2026. Reuters reported that R2 timing had not been determined in June 2025.
When is DeepSeek R2 coming out?
There is no confirmed official DeepSeek R2 release date in the sources checked for this article. Treat any specific R2 date as rumor unless DeepSeek announces it.
Are DeepSeek roadmap rumors reliable?
Some reported claims from reputable outlets may be useful, but they are not official. Community posts, anonymous leaks, and SEO blogs should be treated as low-confidence unless backed by DeepSeek documentation.
Is DeepSeek V4 multimodal?
The checked V4 API docs do not confirm native V4 multimodal image, audio, or video capability. DeepSeek previously hinted at future multimodal support after V3, but that is not the same as confirmed V4 multimodal support.
What DeepSeek models are officially available through the API?
The checked API model list shows deepseek-v4-flash and deepseek-v4-pro. The docs also mention legacy aliases deepseek-chat and deepseek-reasoner, but they are scheduled for deprecation.
What should developers do before older model aliases are deprecated?
Developers should update applications to use deepseek-v4-flash or deepseek-v4-pro, test Thinking and Non-Thinking modes, update pricing assumptions, and keep a fallback path before July 24, 2026.
Where should I check official DeepSeek release news?
Start with DeepSeek API Docs News, the Change Log, Models & Pricing, the /models documentation, and the verified DeepSeek Hugging Face organization.
Is DeepSeek V5 confirmed?
No. No official DeepSeek V5 release or roadmap announcement was found in the checked official sources as of May 7, 2026.
Conclusion
The confirmed DeepSeek Roadmap is best understood through releases, not a classic public roadmap. The major confirmed 2026 milestone is DeepSeek-V4 Preview, with V4-Pro, V4-Flash, 1M context, open weights, API access, and clear migration instructions.
The uncertain side is equally important. DeepSeek R2, V4 final GA timing, native V4 multimodal support, and DeepSeek V5 should not be treated as confirmed unless DeepSeek announces them through official channels. Reputable reporting can be cited as reporting, but it should not be rewritten as fact.
For developers, the action plan is practical: test V4-Pro and V4-Flash, migrate away from legacy aliases, monitor pricing, and track the official Change Log. For readers and AI watchers, the best rule is simple: trust official DeepSeek docs first, reputable reporting second, and social rumors last.
