China AI Hub
Models / DeepSeek-V4-Pro
DeepSeek-V4-Pro
DeepSeek-V4 family · Provider: deepseek · Status: deprecated · Released: 2026-04-24 (V4 Preview) / 2026-08-13 (GA)
DeepSeek-V4-Pro is a DeepSeek-V4 model developed by deepseek, a Chinese AI company headquartered in Hangzhou, Zhejiang, China (derived from the official footer company name 杭州深度求索人工智能基础技术研究有限公司 and Zhejiang ICP / Hangzhou public-security filings, released 2026-04-24 (V4 Preview) / 2026-08-13 (GA) with a 1,048,576-token context window and open weights.
| Key fact | Value |
|---|---|
| Model ID | deepseek-v4-pro |
| Version | 0813 |
| Architecture | MoE: 1.6T total / 49B active parameters |
| Context window | 1,048,576 tokens |
| Max output | 393,216 tokens |
| Open weights | Yes |
| License | MIT |
| Self-hosting | Yes |
| API available | Yes |
| API pricing | $0.66 input / $1.98 output per 1M tokens (USD) · provider pricing page |
| Regions | unknown |
| Cloud providers | DeepSeek Platform |
Capabilities
| Capability | Supported |
|---|---|
| Reasoning | Yes |
| Coding | Yes |
| Math | Yes |
| Chinese | Unknown |
| English | Unknown |
| Multilingual | Unknown |
| Vision | No |
| Audio | Unknown |
| Video | Unknown |
| Tool calling | Yes |
| Function calling | Yes |
| Structured output | Yes |
| Agent capability | Unknown |
| RAG | Unknown |
| Computer use | Unknown |
Benchmark results
| Benchmark | Version | Score | Metric | Date | Source type | Source |
|---|---|---|---|---|---|---|
| HLE | — | 42.7 (60.0 with tools) | accuracy | 2026-08-13 | vendor_reported | link |
| Terminal-Bench 2.1 | — | 87.9 | accuracy | 2026-08-13 | vendor_reported | link |
| DeepSWE | — | 62.7 | accuracy | 2026-08-13 | vendor_reported | link |
| Agents' Last Exam | — | 25.7 | accuracy | 2026-08-13 | vendor_reported | link |
Benchmark scores are single data points, not universal rankings. Vendor-reported scores are labeled as such.
Known limitations
- Deprecation announced 2026-09-10: news page says V4-Pro requests will route to V4.1-Flash after 2026-09-14 until V4.1-Pro launches, but the same-day change log says V4-Pro API service continues with unchanged billing - the official pages conflict
- Vision not supported
- Open-weight HF checkpoint last modified 2026-06-22; unclear whether it matches the 0813 GA checkpoint
- Pricing is peak/off-peak: listed prices are off-peak; peak is 2x
DeepSeek-V4-Pro is a 1.6T-parameter MoE (49B active) flagship with MIT open weights, live on the API since the V4 Preview (2026-04-24) and updated to the 0813 GA checkpoint on 2026-08-13. It supports thinking (default) and non-thinking modes with reasoning_effort low/high/max, 1M-token context, 384K maximum output, JSON output and tool calls; vision is not supported.
DeepSeek positions it as open-source SOTA in agentic coding and leading open models in world knowledge, math/STEM and coding. Deprecation was announced on 2026-09-10 as the V4.1 family rolls out; official pages conflict on whether V4-Pro API service continues after 2026-09-14 (change log) or is routed to V4.1-Flash (news page). Off-peak pricing: $0.66 input / $1.98 output per 1M tokens (as of 2026-09-20).
Provider
Pricing
Benchmarks with results for this model
Agents built on this model
API
Data interpretation
| Field | Value | Evidence type |
|---|---|---|
| Context window | 1,048,576 tokens | Official |
| Architecture | MoE: 1.6T total / 49B active parameters | Official |
| Open weights | Yes | Official |
| License | MIT | Official |
| API pricing | $0.66 / $1.98 per 1M tokens (USD) | Official |
| HLE | 42.7 (60.0 with tools) accuracy | Vendor-reported |
| Terminal-Bench 2.1 | 87.9 accuracy | Vendor-reported |
| DeepSWE | 62.7 accuracy | Vendor-reported |
| Agents' Last Exam | 25.7 accuracy | Vendor-reported |
| Family position | 1 model in the DeepSeek-V4 family | China AI Hub analysis |
Evidence types: Official = vendor documentation, pricing pages or model cards. Vendor-reported = benchmark scores published by the vendor. China AI Hub analysis = derived from the database itself. See the sourcing policy.
Sources
Confidence and source hierarchy per the sourcing policy. Facts change; check the source before relying on this page.
What is the context window of DeepSeek-V4-Pro?
DeepSeek-V4-Pro has a 1,048,576-token context window and a maximum output of 393,216 tokens.
Is DeepSeek-V4-Pro open weight?
Yes — DeepSeek-V4-Pro weights are openly available under the MIT license.
How much does DeepSeek-V4-Pro cost through the API?
$0.66 per 1M input tokens and $1.98 per 1M output tokens (USD).
Where does China AI Hub get its DeepSeek-V4-Pro data?
From 4 sources (official pages first), last verified 2026-09-20. Benchmark scores are labeled by source type; see the sourcing policy for details.