DeepSeek
DeepSeek develops foundation models and operates consumer chat and developer APIs. The current pricing page identifies Flash1017 under deepseek-flash and Pro0813 under deepseek-v4-pro. Older Flash IDs are retired compatibility aliases; historical notices do not establish the current destination of a request.
Readers need one stable company page that connects DeepSeek's chat product, API, open-weight releases, agent integrations, new vision surface, and reported infrastructure expansion without merging model identities or turning a media report into an official commitment.
Checked October 28, 2026 against the linked primary contracts and exact release records. Earlier release chapters retain their dated scope; publisher changes are not independent runtime validation.
- Understand DeepSeek as a company across chat, API, research, and open-weight releases.
- Navigate to current DeepSeek V4 model and agent-integration pages.
- Evaluate infrastructure reports without treating them as product-release facts.
- Separate company-level developments from model IDs, prices, and API availability.
DeepSeek’s official surfaces connect consumer chat, developer APIs and model repositories. Current Flash is DeepSeek-V4.1-Flash-1017; the current pricing page separately lists Pro 0813. Historical September notices and older Flash aliases do not override the present model contract.
- Consumer surface: DeepSeek web chat and app links.
- Developer surface: API platform, documentation, pricing, and service status.
- Research surface: technical reports, model repositories, and released weights.
DeepSeek made V4 Pro generally available on app, web, and API on August 13 and introduced peak/off-peak prices on August 16. On August 21 it added `deepseek-v4-flash-vision-exp`, an experimental hosted model for mixed text and image input, plus image support through base64, URLs, and the Files API. Text Flash and Pro remain separate text-only IDs.
- DeepSeek V4 Flash
Text API and open-weight 0731 model with Responses support, Codex integration, and peak/off-peak pricing.
- DeepSeek V4 Flash Vision Exp
Experimental API-only multimodal model for mixed text and image input.
Bloomberg reported that DeepSeek is planning to build a massive data center in Inner Mongolia as it competes with Silicon Valley model developers. DeepSeek's current official site and API documentation do not announce the project. The correct public status is therefore “Bloomberg-reported plan,” not “DeepSeek officially launched or completed a data center.”
The cited Bloomberg post does not by itself establish a one-gigawatt capacity, a precise city, chip supplier, budget, construction milestone, or 2027 completion date. Those details should remain out of the company page until they are supported by the underlying report, a filing, procurement record, or a first-party DeepSeek statement.
A planned data center may matter for future training capacity, inference supply, latency, and cost, but it does not automatically change a current model ID, price, context window, rate limit, or availability status. Use DeepSeek's API documentation and model-specific pages for current production decisions.
The current pricing page identifies deepseek-flash as DeepSeek-V4.1-Flash-1017 and deepseek-v4-pro as DeepSeek-V4-Pro-0813. The older text Flash and Vision Exp IDs are retired compatibility routes to the current Flash model. Search and Fetch are billed separately at $0.75 per 1,000 successful requests; the documented Harness/Codex promotion waives those service calls before November 1, 2026. A temporary waiver is not the permanent token price. Confirm the exact model and billable tool before budgeting.
- Current official source
Read the exact release or provider scope.
- DeepSeek V4 Pro
Long-context V4 hosted API model now mapped to the DeepSeek-V4-Pro-0813 backend.
DeepSeek’s October 20 DeepInfer release exposes an Apache-2.0 offline inference framework for V4.1 Flash, with documented NVIDIA SM100 and Huawei Ascend 950 support. The public engine does not include the full internal cache and online-service infrastructure. This release changes what self-hosters can inspect, not the hosted API model ID, price or quota.
- DeepInfer
Check supported hardware, prerequisites and missing service components.
Text API and open-weight 0731 model with Responses support, Codex integration, and peak/off-peak pricing.
Experimental API-only multimodal model for mixed text and image input.
Long-context V4 hosted API model now mapped to the DeepSeek-V4-Pro-0813 backend.
Third-party terminal coding agent documented as a DeepSeek API integration.
Tesla China in-car AI feature whose localized stack has also been reported to use DeepSeek for some interactions.
Source confidence
DeepSeek API Docs
DeepSeek API Docs
DeepSeek
DeepSeek API Docs
GitHub / deepseek-ai
Bloomberg
Project publisher
DeepSeek
DeepSeek FAQ
Common questions about DeepSeek.
Is DeepSeek building a data center in Inner Mongolia?+
Bloomberg reports that DeepSeek is planning a massive AI data center in Inner Mongolia. DeepSeek has not published a matching first-party announcement on the official surfaces cited here, so treat it as a credible reported plan rather than a confirmed construction or completion milestone.
Does the reported data center change DeepSeek API pricing or models now?+
No current API change follows automatically from the report. Use DeepSeek's official API documentation for model IDs, prices, limits, and availability, and treat infrastructure plans as a separate company-level development.
Is every DeepSeek model open source?+
No single label covers every DeepSeek surface. DeepSeek publishes research, repositories, and model weights, while hosted chat and API products have separate service terms and model-specific licenses. Verify the exact repository, license, and deployment surface for each model.