DeepSeek-V4-Pro-0813 is the official release of DeepSeek-V4-Pro, superseding the preview version, with greatly enhanced agentic capabilities and performance improvements that are especially pronounced in production environments. It is built on the DeepSeek-V4-Pro (Preview) model structure, with a DSpark speculative decoding module attached.
DeepSeek-V4-Pro-0813 can be customized with your data to improve responses. Fireworks uses LoRA to efficiently train and deploy your personalized model
DeepSeek-V4-Pro-0813 is available via Fireworks' serverless API, where you pay per token. There are several ways to call the Fireworks API, including Fireworks' Python client, the REST API, or OpenAI's Python client.
On-demand deployments allow you to use DeepSeek-V4-Pro-0813 on dedicated GPUs with Fireworks' high-performance serving stack with high reliability and no rate limits.
Run queries immediately, pay only for usage
Only on serverless, starting on September 25, 2026. On-demand deployment will continue to be available. Migrate serverless workloads to DeepSeek V4.1 Flash.
The official release of DeepSeek-V4-Pro from DeepSeek-AI, superseding the preview checkpoint and adding a DSpark speculative decoding module. DeepSeek shipped it on 2026-08-13.
Agentic and tool-augmented work, with DeepSeek reporting Terminal Bench 2.1 at 87.9, Cybergym at 83.3, and Toolathlon-Verified at 74.1. Per DeepSeek’s changelog, its API natively accepts the OpenAI Responses API format.
1,048,576 tokens, set as max_position_embeddings in the model config. DeepSeek describes the V4 series as supporting one million tokens.
Fireworks lists a 1040k-token context window, available on both the serverless endpoint and on-demand deployments.
Fireworks serves DeepSeek's mixed-precision checkpoint: MoE expert weights in FP4, remaining weights in block-quantized FP8 (e4m3). No separate 4-bit or 8-bit variant is published.
DeepSeek recommends a temperature of 1.0. For local deployment it recommends top_p of 0.95 for agentic scenarios and 1.0 otherwise.
Fireworks defaults max_tokens to 2048 and supports generation up to the full context window. DeepSeek recommends capping output at 384K tokens at high and max reasoning effort.
Yes to both. Streaming runs through the stream parameter, and function calling is supported. The model also exposes reasoning_effort with three levels: low, high, max.
DeepSeek-V4-Pro-0813 has 1.6 trillion total parameters as listed by Fireworks; Hugging Face's model-size display shows 1.7T.
Yes. Fireworks supports LoRA fine-tuning for DeepSeek-V4-Pro-0813.
Fireworks bills input, cached input, and output tokens separately, priced at $1.32, $0.044, and $3.96 per 1M tokens.
Serverless ceilings for this model's size tier default to 21.6M total prompt TPM, 5.4M uncached prompt TPM, and 216k generated TPM, adaptive per account and model. On-demand deployments carry no rate limits.
Fireworks announces serverless model deprecations in advance, following its serverless model lifecycle policy. On-demand deployments are recommended for long-term version stability.
Released under the MIT License, which permits commercial use and covers both the repository and the model weights.
No. Fireworks operates zero data retention by default and does not log prompt or generation data for open models without opt-in. Only metadata such as token counts is logged.
