Short answer: The official DeepSeek API materials reviewed for this guide do not list an R2 model or establish R2’s official availability, capabilities, direct API access, or price. They do document deepseek-flash and deepseek-v4-pro; the API quick start says deepseek-flash is served by DeepSeek-V4.1-Flash-1017. A third-party listing is not the same as confirmation from DeepSeek. [1][6] (api-docs.deepseek.com)
For a production decision: Don’t build against an assumed deepseek-r2 ID or budget using rumored R2 prices. If you want to evaluate DeepSeek through its direct API, use a model named in its current documentation and test it on your workload. If a gateway advertises R2, verify what that provider routes to, which API contract it supports, and what it charges.
What is DeepSeek R2, and is it available?
R2 has been the subject of release speculation, but older reports do not establish its current status. A SitePoint article projected an August 2025 release and discussed possible architecture, context length, and reasoning or coding improvements. It described these as expectations or rumors, and its forward-looking timeline is stale. An August 14, 2025 SiliconANGLE report said R2 had not launched by that date; that is historical reporting, not evidence of current availability. [7][8]
CometAPI’s comparison page lists the model ID deepseek-r2 and endpoint path /v1/chat/completions. It does not establish that DeepSeek offers R2 through its own API: the listing is a third-party claim, and the page leaves key capability fields blank. [6]
Practical conclusion: Treat R2’s official status, specifications, and direct API access as unconfirmed unless DeepSeek publishes current documentation or an announcement that establishes them. Do not equate a gateway’s model listing with a direct DeepSeek offering.
What can developers evaluate through DeepSeek’s documented API?
DeepSeek’s API quick start names deepseek-flash and deepseek-v4-pro. It says deepseek-flash is served by DeepSeek-V4.1-Flash-1017. DeepSeek’s September 9, 2026 announcement describes V4.1-Flash as live on its API with native multimodal support. [1][4]
The announcement said deepseek-v4-pro requests would route to V4.1-Flash at Flash rates starting September 14, 2026, until V4.1-Pro launches. A later entry in DeepSeek’s changelog says the company decided to continue providing V4 Pro API service after September 14, with billing unchanged. Because those updates differ, don’t assume the originally announced routing change took effect; check the current model and pricing documentation before relying on that route. [4][5]
| If your priority is… | Documented starting point | What to verify in your evaluation |
|---|---|---|
| Testing a model named in DeepSeek’s current API quick start | deepseek-flash |
Quality on your prompts, latency, output behavior, and current pricing. [1][2] |
| Testing the separately named V4 Pro route | deepseek-v4-pro appears in the quick start and pricing documentation |
Confirm its current serving behavior and billing; the announcement and later changelog differ on the planned routing change. [1][2][4][5] |
| Finding an officially documented R2 API option | None identified in the reviewed official materials | Wait for an official model ID and API contract, or treat a third-party offer as that provider’s product. [1][3] |
DeepSeek documents GET https://api.deepseek.com/models for listing currently available models. The response schema describes fields such as model ID, context window, maximum output, input and output modalities, effort levels, and API capabilities. Some metadata may be omitted when unavailable. [3]
Direct API example: test a documented model, not an assumed R2 ID
DeepSeek’s official quick start provides the chat-completions request below: it uses POST to https://api.deepseek.com/chat/completions, JSON content, Bearer-token authentication, the deepseek-flash model, and a messages array. This is a documented example for deepseek-flash, not an R2 request. Set DEEPSEEK_API_KEY to your DeepSeek API key before running it. (api-docs.deepseek.com)
curl https://api.deepseek.com/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer ${DEEPSEEK_API_KEY}" -d '{"model":"deepseek-flash","messages":[{"role":"system","content":"You are a helpful assistant."},{"role":"user","content":"Hello!"}],"thinking":{"type":"enabled"},"reasoning_effort":"high","stream":false}'
The quick start describes this as a non-streaming example and says you can set stream to true to receive a stream. The example does not establish R2 compatibility or production suitability. Confirm the current model identifier, account limits, error handling, and response handling before deployment. [1]
If a gateway advertises R2, confirm with that provider the full base URL, model ID, authentication method, request and response format, streaming support, limits, and upstream model. Do not combine a gateway’s endpoint or pricing with DeepSeek’s direct-API instructions.
DeepSeek R2 pricing: what can be used for budgeting?
No current official R2 price was confirmed in the reviewed DeepSeek pricing documentation. SitePoint’s estimate of approximately $0.55–$1.00 per million input tokens was a projection, not an official R2 rate. CometAPI displays $75.00 / M tokens and $93.75 / M tokens alongside a -20% label for input and output; the page does not make the price relationship clear. Neither source is a reliable basis for a direct-API R2 budget. [6][7]
For a documented direct-API model, DeepSeek’s pricing page lists deepseek-flash rates per 1 million tokens: [2]
| Token type | Peak price | Off-peak price |
|---|---|---|
| Input, cache miss | $0.30 / 1M tokens | $0.15 / 1M tokens |
| Input, cache hit | $0.006 / 1M tokens | $0.003 / 1M tokens |
| Output | $1.20 / 1M tokens | $0.60 / 1M tokens |
The pricing page says off-peak rates are half the peak rates. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, excluding Chinese public holidays; all other hours are off-peak. The page displays a dollar sign but does not spell out a currency denomination in the pricing text. DeepSeek says prices may change, so consult its current pricing page when estimating costs. [2]
Worked example: At peak rates, 1 million cache-miss input tokens plus 100,000 output tokens totals about $0.42 using the displayed dollar amounts: $0.30 + $0.12. This is an arithmetic example, not an R2 estimate; cache-hit and off-peak usage would change the calculation. [2]
How to decide whether to adopt a model
The reviewed evidence does not establish an R2 result that supports recommending it over an incumbent model. Evaluate a currently available model against your own requirements, and revisit R2 if DeepSeek publishes an official model identity and product details.
- Choose representative tasks. Save prompts that reflect real work, including difficult or failure-prone cases.
- Keep a baseline. Run the same prompts through your current model and record outputs and latency.
- Score what matters. Use a task-specific quality rubric; record latency, errors, refusals, and token usage. DeepSeek’s published prices distinguish input cache hits, cache misses, and output, so track those categories separately when estimating direct API cost. [2]
- Compare like with like. Keep prompts and test conditions consistent. Don’t substitute projected benchmarks or rumored R2 limits for results on your own workload.
- Roll out only after meeting thresholds. If a candidate meets your quality, latency, and cost requirements, start with a controlled rollout and keep a fallback. Provider abstraction and staged rollout are general engineering practices, not guarantees of R2 compatibility.
Bottom line
The reviewed official DeepSeek materials document V4-series API models but do not establish an official R2 model ID, R2 capabilities, direct R2 API access, or R2 pricing. For a direct-API test, use a model named in the current documentation, such as deepseek-flash. For an R2 gateway offer, verify that provider’s model routing and terms independently. Treat R2 specifications and prices from older projections or third-party listings as unconfirmed until supported by current authoritative documentation. [1][2][3][6][7]
Sources
Your First API Call | DeepSeek API Docs
Models & Pricing | DeepSeek API Docs
Lists Models | DeepSeek API Docs
Introducing DeepSeek-V4.1-Flash | DeepSeek
Change Log | DeepSeek API Docs
DeepSeek R2 vs AUTO | CometAPI
DeepSeek R2: What Developers Need to Know Before August
DeepSeek R2 model release reportedly held back by faulty Huawei chips