DeepSeek-V4-Flash API Now in Public Beta

DeepSeek released the V4-Flash API in public beta on July 31, 2026. The model name is deepseek-v4-flash. The API calling method remains unchanged. Set the model parameter to deepseek-v4-flash to use the latest version.

This update only affects the V4-Flash API. The V4-Pro API and the APP/WEB models remain unchanged. The official V4-Pro release will follow soon.

Benchmark Results: V4-Flash vs V4-Pro-Preview

DeepSeek reports that V4-Flash significantly outperforms V4-Pro-Preview on agent benchmarks:

  • Terminal Bench 2.1: 82.7
  • NL2Repo: 54.2
  • Cybergym: 76.7
  • DeepSWE: 54.4
  • Toolathlon verified: 70.3
  • Agent Last Exam: 25.2
  • Automation Bench (Public): 25.1
  • DSBench-FullStack: 68.7
  • DSBench-Hard: 59.6

These are internal or public benchmark scores. The public benchmark tests used the DeepSeek Harness minimal mode (to be released soon) with max effort, topp=0.95, and temperature=1.0.

DSBench-FullStack and DSBench-Hard are internal test sets. FullStack tests full-stack development. Hard tests coding agent hard problems.

Architecture and Compatibility

DeepSeek-V4-Flash-0731 keeps the same model architecture and size as the V4-Flash-Preview. It was only re-post-trained.

The official V4-Flash natively supports the Responses API format. It is specifically adapted for Codex. For configuration details, refer to the documentation.

Historical Context: The Evolution of DeepSeek's API

DeepSeek's API has evolved rapidly. On April 24, 2026, the API added V4-Pro and V4-Flash. They were accessible via both OpenAI ChatCompletions and Anthropic interfaces. The legacy model names deepseek-chat and deepseek-reasoner were scheduled for discontinuation on July 24, 2026. During the transition, they pointed to V4-Flash's non-thinking and thinking modes.

Earlier, on December 1, 2025, both deepseek-chat and deepseek-reasoner were upgraded to DeepSeek-V3.2. There was also a temporary DeepSeek-V3.2-Speciale endpoint, expiring December 15, 2025.

In September 2025, DeepSeek released V3.2-Exp and V3.1-Terminus. V3.1-Terminus addressed language consistency and agent capabilities. V3.1, released August 21, 2025, introduced hybrid reasoning architecture and improved agent capabilities, scoring 66.0 on SWE-bench Verified.

In May 2025, deepseek-reasoner was upgraded to R1-0528, with significant benchmark improvements: AIME 2025 from 70.0 to 87.5, GPQA from 71.5 to 81.0, LCB_v6 from 63.5 to 73.3, Aider from 57.0 to 71.6.

In March 2025, deepseek-chat was upgraded to V3-0324, with MMLU-Pro from 75.9 to 81.2 and AIME from 39.6 to 59.4.

What This Means for Developers

You can start using V4-Flash immediately. The API is backward compatible with the OpenAI interface. If you were using deepseek-chat, switch to deepseek-v4-flash.

from openai import OpenAI

client = OpenAI(
    api_key="your-api-key",
    base_url="https://api.deepseek.com"
)

response = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Write a Python function to flatten a nested list."}]
)
print(response.choices[0].message.content)

For Codex integration, check the docs for the Responses API format.

The Bottom Line

DeepSeek-V4-Flash is a major upgrade for agentic workflows. The benchmark scores are impressive. The API is ready for production beta. Start testing now.