DeepSeek V4 Pro 0813 quietly released
TL;DR Highlight
Change base_url and call DeepSeek models through the OpenAI Responses API format
Who Should Read
Backend developers with Responses API clients or Codex setups who want to try DeepSeek models
Core Mechanics
- Responses API support exists for Codex demand
- Set base_url to https://api.deepseek.com.
- The docs say a simple configuration lets Codex use DeepSeek models.
- Two models: deepseek-flash and deepseek-v4-pro
- The doc examples call deepseek-flash.
- Image input is accepted with deepseek-flash.
- Server-side web_search is v4-pro only
- With deepseek-v4-pro the API searches on the server and adds results to context.
- With deepseek-flash, web_search is ignored.
- The API is stateless
- previous_response_id, conversation, store and background are not supported.
- Responses always carry store: false.
- Unsupported parameters are silently ignored
- So existing Responses API clients can connect without modification.
Community Reactions
- Two of the top three HN links were about this
- One commenter joked that it was not so quiet after all.
- Opinion that benchmarks may show nothing dramatic
- A commenter speculated Flash already replaces Opus 4.5 for local agentic coding, so Pro may lack a similar win.
- Pricing page link shared
- A commenter pointed to the pricing docs for per-model costs.
How to Apply
- Swap base_url in existing OpenAI SDK code and call once
- Today, set base_url to https://api.deepseek.com and model to deepseek-flash, then call responses.create.
- Print response.output_text plus cached_tokens and reasoning_tokens from usage.
- Audit previous_response_id usage and manage history yourself
- The server keeps no state, so send prior turns in input each request.
- Find and remove logic relying on ignored parameters like store or background.
- Route web-search paths to deepseek-v4-pro
- Add {"type": "web_search"} to tools and set model to deepseek-v4-pro to test.
- Flash ignores search, so make the model explicit in your routing rule.
Try It
# Run in a terminal (pip3 install openai)
from openai import OpenAI
client = OpenAI(api_key="<your DeepSeek API Key>", base_url="https://api.deepseek.com")
stream = client.responses.create(
model="deepseek-flash",
instructions="You are a helpful assistant.",
input="Hi, how are you?",
stream=True,
)
for event in stream:
if event.type == "response.output_text.delta":
print(event.delta, end="")
# Next: drop stream=True and print response.usage to check cached_tokensTerminology
Related Papers
Same Simulation, Different Evidence: Where Retrieved Tool Output Costs LLM Agents Accuracy
Saving long tool output to files cut GPT-5.1 accuracy by 11.3 points; naming the authoritative file restored it.
Fake News, Real Crashes: Information Injection Attacks on LLM Financial Trading Agents
Fake headlines hijacked LLM trading agents 62% of the time versus 14% for prompt injection.
Harness Engineering 101: The LLM is the brain, the harness is the body
An agent is a loop managing a JSON array; 16 chapters build one from scratch
Harness2 by Google Research – Recursive Agent Harnessing
Recursively refining the harness with fixed model weights raised benchmark scores by up to 11.8 points.
Ousted Savannah police officer gave ICE agent husband access to Flock system
A police officer shared her Flock login with her ICE-agent husband; an audit found 983 searches and she was fired
Tmux and Claude Code: Sessions, Send-Keys, and the Ceiling
Running Claude Code agents in tmux works well, but limits appear around four agents.