Model API

DeepSeek

DeepSeek · source page ↗ · last checked Jul 6, 2026, 12:00 AM

Current pricing

captured Jul 6, 2026, 12:00 AM

Model / item	Input	Output
deepseek-v4-flash	$0.14 / 1M tokens (cache miss)	$0.28 / 1M tokens · Cache-hit input: $0.0028 / 1M tokens. 1M context, 384K max output. Non-thinking + thinking (default) modes. Replaces deprecated deepseek-chat (non-thinking) / deepseek-reasoner (thinking).
deepseek-v4-pro	$0.435 / 1M tokens (cache miss)	$0.87 / 1M tokens · Cache-hit input: $0.003625 / 1M tokens. 1M context, 384K max output. Non-thinking + thinking (default) modes.

USD. Billing: expense = number of tokens x price. "Product prices may vary and DeepSeek reserves the right to adjust them" — check the page for the latest. Legacy model names deepseek-chat and deepseek-reasoner are deprecated 2026/07/24 15:59 UTC (mapping to v4-flash non-thinking / thinking modes). No off-peak/discount tier listed on this page version.

Curated snapshot · as of 2026-06-15. Live diffs are recorded in the change history below.

Rate & usage limits structure only

This provider doesn't publish numeric request/token limits on its docs — only the usage-tier structure below. Your actual limits appear in your account dashboard.

Tier	Notes
Default account — deepseek-v4-pro (flagship)	Concurrency Limit: 500 concurrent connections for model deepseek-v4-pro (the flagship/pro model). Calculated at account level across all API Keys. Exceeding it returns HTTP 429. No requests-per-minute, tokens-per-minute, or requests-per-day limits are published; DeepSeek limits by concurrency only.
Default account — deepseek-v4-flash	Concurrency Limit: 2500 concurrent connections for model deepseek-v4-flash. Account-level, across all API Keys; HTTP 429 when exceeded. No RPM/TPM/RPD published.
Increased-concurrency quota (capacity expansion) — per user_id	For accounts granted increased concurrency, a per-user_id concurrency limit also applies: 500 concurrent for deepseek-v4-pro and 2500 concurrent for deepseek-v4-flash per user_id (an empty id is treated as a special user_id). Exceeding a user_id's limit returns HTTP 429 for that user_id. Capacity expansion is free — submit a capacity-expansion request; concurrency matched to actual business needs. user_id must match regex [a-zA-Z0-9\-_]+, max length 512.

DeepSeek does NOT publish RPM/TPM/RPD rate limits. The API is constrained by CONCURRENCY (concurrent connections), not request/token rate. Limits are per-model and calculated at the account level regardless of which API Key is used. A request counts as one concurrent connection from when it is sent until the model response is complete; exceeding the limit returns HTTP 429. Concurrency can be raised for free via a capacity-expansion request ("There is no additional cost for capacity expansion"). If a request has not started inference after 10 minutes the server closes the connection. There are no named account tiers (Free/Tier 1/etc.) on this page — the only distinction is base accounts vs accounts with an increased-concurrency quota (which adds per-user_id limits).

Source: https://api-docs.deepseek.com/quick_start/rate_limit

Curated snapshot · as of 2026-06-15. Auto-refreshes once the limits page changes.

Currently listed models

captured Jul 6, 2026, 12:00 AM

deepseek-ai/DeepSeek-Coder-V2-Base license:other 2024-06-14
deepseek-ai/DeepSeek-Coder-V2-Instruct license:other 2024-06-14
deepseek-ai/DeepSeek-Coder-V2-Instruct-0724 license:other 2024-09-05
deepseek-ai/DeepSeek-Coder-V2-Lite-Base license:other 2024-06-14
deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct license:other 2024-06-14
deepseek-ai/DeepSeek-Math-V2 license:apache-2.0 2025-11-27
deepseek-ai/DeepSeek-OCR license:mit 2025-10-17
deepseek-ai/DeepSeek-OCR-2 license:apache-2.0 2026-01-27
deepseek-ai/DeepSeek-Prover-V2-671B license:? 2025-04-30
deepseek-ai/DeepSeek-R1 license:mit 2025-01-20
deepseek-ai/DeepSeek-R1-0528 license:mit 2025-05-28
deepseek-ai/DeepSeek-R1-0528-Qwen3-8B license:mit 2025-05-29
deepseek-ai/DeepSeek-R1-Distill-Llama-70B license:mit 2025-01-20
deepseek-ai/DeepSeek-R1-Distill-Llama-8B license:mit 2025-01-20
deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B license:mit 2025-01-20
deepseek-ai/DeepSeek-R1-Distill-Qwen-32B license:mit 2025-01-20
deepseek-ai/DeepSeek-R1-Distill-Qwen-7B license:mit 2025-01-20
deepseek-ai/DeepSeek-R1-Zero license:mit 2025-01-20
deepseek-ai/DeepSeek-V2 license:other 2024-04-22
deepseek-ai/DeepSeek-V2-Chat license:other 2024-04-28
deepseek-ai/DeepSeek-V2-Chat-0628 license:other 2024-07-18
deepseek-ai/DeepSeek-V2-Lite license:other 2024-05-15
deepseek-ai/DeepSeek-V2-Lite-Chat license:other 2024-05-15
deepseek-ai/DeepSeek-V2.5 license:other 2024-09-05
deepseek-ai/DeepSeek-V2.5-1210 license:other 2024-12-10
deepseek-ai/DeepSeek-V3 license:? 2024-12-25
deepseek-ai/DeepSeek-V3-0324 license:mit 2025-03-24
deepseek-ai/DeepSeek-V3.1 license:mit 2025-08-21
deepseek-ai/DeepSeek-V3.1-Base license:mit 2025-08-19
deepseek-ai/DeepSeek-V3.1-Terminus license:mit 2025-09-22
deepseek-ai/DeepSeek-V3.2 license:mit 2025-12-01
deepseek-ai/DeepSeek-V3.2-Exp license:mit 2025-09-29
deepseek-ai/DeepSeek-V3.2-Exp-Base license:mit 2025-09-29
deepseek-ai/DeepSeek-V3.2-Speciale license:mit 2025-11-28
deepseek-ai/DeepSeek-V4-Flash license:mit 2026-04-22
deepseek-ai/DeepSeek-V4-Flash-DSpark license:mit 2026-06-27
deepseek-ai/DeepSeek-V4-Pro license:mit 2026-04-22
deepseek-ai/DeepSeek-V4-Pro-DSpark license:mit 2026-06-27
deepseek-ai/ESFT-gate-code-lite license:? 2024-07-04
deepseek-ai/ESFT-gate-intent-lite license:? 2024-07-04
deepseek-ai/ESFT-gate-law-lite license:? 2024-07-04
deepseek-ai/ESFT-gate-math-lite license:? 2024-07-04
deepseek-ai/ESFT-gate-summary-lite license:? 2024-07-04
deepseek-ai/ESFT-gate-translation-lite license:? 2024-07-04
deepseek-ai/ESFT-token-code-lite license:? 2024-07-04
deepseek-ai/ESFT-token-intent-lite license:? 2024-07-04
deepseek-ai/ESFT-token-law-lite license:? 2024-07-04
deepseek-ai/ESFT-token-math-lite license:? 2024-07-04
deepseek-ai/ESFT-token-summary-lite license:? 2024-07-04
deepseek-ai/ESFT-token-translation-lite license:? 2024-07-04
deepseek-ai/ESFT-vanilla-lite license:? 2024-07-04
deepseek-ai/Janus-1.3B license:mit 2024-10-18
deepseek-ai/Janus-Pro-1B license:mit 2025-01-26
deepseek-ai/Janus-Pro-7B license:mit 2025-01-26
deepseek-ai/JanusFlow-1.3B license:mit 2024-11-12
deepseek-ai/deepseek-coder-1.3b-base license:other 2023-10-28
deepseek-ai/deepseek-coder-1.3b-instruct license:other 2023-10-29
deepseek-ai/deepseek-coder-33b-base license:other 2023-10-28
deepseek-ai/deepseek-coder-33b-instruct license:other 2023-11-01
deepseek-ai/deepseek-coder-5.7bmqa-base license:other 2023-10-31

In the news · DeepSeek

Tsinghua Special Scholarship Winner Gu Yuxian Joins DeepSeek Ahead of V4 Launch via Pandaily — Tsinghua Special Scholarship Winner Gu Yuxian Joins DeepSeek Ahead of V4 Launch Pandaily · Jul 5, 2026, 9:01 AM
How to Use the DeepSeek API for Free (or Near-Free) via openPR.com — How to Use the DeepSeek API for Free (or Near-Free) openPR.com · Jul 4, 2026, 6:15 AM
DeepSeek Hikes API Prices 2x During Peak Hours; Goldman Sachs Says China's AI Model Price War Nearing End via finance.biggo.com — DeepSeek Hikes API Prices 2x During Peak Hours; Goldman Sachs Says China's AI Model Price War Nearing End finance.biggo.com · Jul 3, 2026, 3:25 PM
DeepSeek V4 To Add Peak-Hour Pricing to Its API via WinBuzzer — DeepSeek V4 To Add Peak-Hour Pricing to Its API WinBuzzer · Jul 3, 2026, 9:34 AM
Breach Roundup: DeepSeek Sparks Browser Ransomware via BankInfoSecurity — Breach Roundup: DeepSeek Sparks Browser Ransomware BankInfoSecurity · Jul 2, 2026, 9:00 PM
Digest: Apple & EU Hold Siri AI Talks; DeepSeek Introduces Peak-Hour API Pricing via exchangewire.com — Digest: Apple & EU Hold Siri AI Talks; DeepSeek Introduces Peak-Hour API Pricing exchangewire.com · Jul 2, 2026, 8:55 AM
Together AI Raises $800M to Scale Open Models via Let's Data Science — Together AI Raises $800M to Scale Open Models Let's Data Science · Jul 1, 2026, 6:39 PM
DeepSeek Raises $7B in Record-Breaking China AI Funding Round, Targets Domestic Computing Infrastructure via Pandaily — DeepSeek Raises $7B in Record-Breaking China AI Funding Round, Targets Domestic Computing Infrastructure Pandaily · Jul 1, 2026, 8:10 AM

Importance-filtered press coverage (Google News) mentioning DeepSeek. Headlines link to the original; verify before acting.

Change history

Model Jun 28, 2026, 12:47 PM

DeepSeek (weights) models changed

Removed deepseek-coder-6.7b-base model from available weights.
- deepseek-coder-6.7b-base model removed
- license:other
- dated 2023-10-23
View raw diff +0 −1
```
- deepseek-ai/deepseek-coder-6.7b-base license:other 2023-10-23
```
Model Jun 27, 2026, 6:00 AM

DeepSeek (weights) models changed

Two new DeepSeek models added: DeepSeek-V4-Flash-DSpark and DeepSeek-V4-Pro-DSpark, both MIT licensed with release date 2026-06-27.
- deepseek-ai/DeepSeek-V4-Flash-DSpark
- deepseek-ai/DeepSeek-V4-Pro-DSpark
- license:mit, 2026-06-27
View raw diff +2 −0
```
+ deepseek-ai/DeepSeek-V4-Flash-DSpark license:mit 2026-06-27
+ deepseek-ai/DeepSeek-V4-Pro-DSpark license:mit 2026-06-27
```