Identity
- Status
- active
- Developer ID
deepseek-ai/DeepSeek-V4-Pro-0813- Released
- Aug 13, 2026
- Version
- DeepSeek-V4-Pro-0813
- License
- mit
DeepSeekDeepSeek-V4-Pro-0813
DeepSeek-V4-Pro-0813 is DeepSeek's open-weight production release of V4 Pro, with a 1M-token context window, three reasoning-effort levels, and a DSpark speculative-decoding module.
- Context
- 1,048,576 tokens
- Inputs
- Text
- Outputs
- Text
- Released
- Aug 13, 2026
Providers & Pricing
Estimate workload cost →| Provider | Input / 1M | Output / 1M | Cache read / 1M | Cache write / 1M | Observed | Evidence |
|---|---|---|---|---|---|---|
 DeepinfraAvailable | $1.30 | $2.60 | $0.10 | — | Sep 3, 2026 | Provider-reported |
Benchmarks
All benchmark results →| ARC-AGI-1verified-v1-b2e2a31b3d53 | 90.50 | — | — | DeepSeek V4 Pro 0813 (Low)unknown | Original sourcewinner eligible | Third-party benchmark |
| ARC-AGI-2verified-v2-326661568d5f | 61.25 | — | — | DeepSeek V4 Pro 0813 (Max)unknown | Original sourcewinner eligible | Third-party benchmark |
| ToneBench2026-08-28-10-task-cd9819ab6e4d | 84.04 | 18,922 | 50 | DeepSeek V4 Pro 0813 (max)average per case | Original sourcewinner eligible | Third-party benchmark |
| ToneBench2026-08-28-10-task-cd9819ab6e4d | 79.67 | 6,473 | 50 | DeepSeek V4 Pro 0813 (default)average per case | Original sourcewinner eligible | Third-party benchmark |
At a Glance
Model Facts
Capacity
- Context
- 1,048,576 tokens
- Maximum output
- Unknown
- Total parameters
- 1,598,839,674,782
- Active parameters
- 49,000,000,000
- Knowledge cutoff
- Not reported
Interface and Access
- Inputs
- Text
- Outputs
- Text
- API available
- Unknown
- Open weights
- Yes
- Self-hostable
- Yes
- Reasoning
- Yes
- Vision
- No
- Tool calling
- Yes
- Capabilities
- agents, chat, coding, generation, reasoning, tools
This page represents one immutable developer model ID. Serving endpoints, pricing, regions, quantizations, and service tiers attach separately and never create duplicate model records.
Version Lineage
PredecessorUnknown
SuccessorsUnknown
Aliasesdeepseek-ai/DeepSeek-V4-Pro-0813, deepseek-v4-pro-0813, DeepSeek V4 Pro 0813
Explicit Unknowns
- Maximum output
- Knowledge cutoff
- API availability
Primary Evidence
Sources and Observation Date
deepseek-ai/DeepSeek-V4-Pro-0813 developer model repositoryDeepSeek · observed Sep 3, 2026deepseek-ai/DeepSeek-V4-Pro-0813 README.md at 72e1d3230f6cDeepSeek · observed Sep 3, 2026deepseek-ai/DeepSeek-V4-Pro-0813 config.json at 72e1d3230f6cDeepSeek · observed Sep 3, 2026deepseek-ai/DeepSeek-V4-Pro-0813 immutable Hub API metadataDeepSeek · observed Sep 3, 2026deepseek-ai/DeepSeek-V4-Pro-0813 tokenizer_config.json at 72e1d3230f6cDeepSeek · observed Sep 3, 2026DeepInfra OpenAI-compatible model catalogDeepInfra · observed Sep 2, 2026Fireworks Ai catalogFireworks Ai · observed Sep 3, 2026Together Ai catalogTogether Ai · observed Sep 3, 2026
Comparable Peers
Related Model Comparisons
| A | Pair | B | Context |
|---|---|---|---|
DeepSeek-V4-Pro-0813DeepSeek | vs |  Kimi K2.7 CodeMoonshot AI | cross-developer peerstext |
DeepSeek-V4-Pro-0813DeepSeek | vs | DeepSeek-V4-ProDeepSeek | family variantstext |
DeepSeek-V4-Pro-0813DeepSeek | vs | DeepSeek-V4-FlashDeepSeek | family variantstext |
DeepSeek-V4-Pro-0813DeepSeek | vs | Kimi-K3Moonshot AI | cross-developer peerstext |
DeepSeek-V4-Pro-0813DeepSeek | vs | cross-developer peerstext | |
DeepSeek-V4-Pro-0813DeepSeek | vs |  Qwen3.8-MaxQwen | cross-developer peerstext |
DeepSeek-V4-Pro-0813DeepSeek | vs | Qwen3.8-FlashQwen | cross-developer peerstext |
DeepSeek-V4-Pro-0813DeepSeek | vs |  GLM-5.3Z.ai | cross-developer peerstext |
DeepSeek-V4-Pro-0813DeepSeek | vs | DeepSeek-V4-Flash-Vision-ExpDeepSeek | family variantstext |
DeepSeek-V4-Pro-0813DeepSeek | vs | cross-developer peerstext | |
DeepSeek-V4-Pro-0813DeepSeek | vs | DeepSeek-V4-Flash-BaseDeepSeek | family variantstext |
DeepSeek-V4-Pro-0813DeepSeek | vs | GLM-5.2Z.ai | cross-developer peerstext |
Questions
DeepSeek-V4-Pro-0813 FAQs
What is DeepSeek-V4-Pro-0813?+
DeepSeek-V4-Pro-0813 is DeepSeek's open-weight production release of V4 Pro, with a 1M-token context window, three reasoning-effort levels, and a DSpark speculative-decoding module. It is developed by DeepSeek and its current sourced lifecycle status is active.
What inputs and outputs does DeepSeek-V4-Pro-0813 support?+
DeepSeek-V4-Pro-0813's current record lists Text as input and Text as output.
How much context does DeepSeek-V4-Pro-0813 support?+
The current record lists a 1,048,576-token context window and does not report a maximum output length.
Is DeepSeek-V4-Pro-0813 available through an API?+
DeepSeek-V4-Pro-0813's current primary-source record does not establish API availability. No provider endpoint is treated as available without direct provider evidence.
How large is DeepSeek-V4-Pro-0813?+
The current record lists 1,598,839,674,782 total parameters and 49,000,000,000 active parameters. Its architecture is DeepseekV4ForCausalLM.
What capabilities and tasks does DeepSeek-V4-Pro-0813 support?+
DeepSeek-V4-Pro-0813's recorded capabilities are agents, chat, coding, generation, reasoning, and tools. Its supported tasks are coding, reasoning, and text.
How much does DeepSeek-V4-Pro-0813 cost through an API?+
The lowest directly sourced prices currently attached to DeepSeek-V4-Pro-0813 are $1.30 per million input tokens through Deepinfra and $2.60 per million output tokens through Deepinfra. Prices are provider-specific and should be checked against each cited observation date.
Can DeepSeek-V4-Pro-0813 be self-hosted?+
DeepSeek-V4-Pro-0813 is an open-weight model. Self-hosting is supported, and the recorded license is mit.
Does DeepSeek-V4-Pro-0813 have related versions or aliases?+
No predecessor is recorded. No successor is recorded. Recorded aliases are deepseek-ai/DeepSeek-V4-Pro-0813, deepseek-v4-pro-0813, and DeepSeek V4 Pro 0813.
Which models can DeepSeek-V4-Pro-0813 be compared with?+
The current compatible peer set includes Kimi K2.7 Code, DeepSeek-V4-Pro, DeepSeek-V4-Flash, Kimi-K3, and NVIDIA Nemotron 3.5 Lightning 30B-A3B. Compatibility requires shared sourced modalities and tasks; it does not imply that one model is better. Browse model comparisons →