DeepSeek V4 Flash 0731

DeepSeek V4 Flash 0731 is the official release of V4 Flash, re-post-trained from the preview build rather than re-pretrained. The architecture, size, 1M-token context and per-token price are all unchanged, but agentic and coding performance moved sharply: it scores 52 on the Artificial Analysis Intelligence Index against 40 for the preview, gains 17 points on Terminal-Bench 2.1, and cuts its hallucination rate by 12 points. Knowledge benchmarks barely shifted, so the gain is in doing rather than knowing. Reach for it on tool-use loops, coding agents, and long-running automation where the preview fell short. The preview build stays available under the original id for anyone who needs the exact behaviour they built against.

Quick start

Use any OpenAI-compatible client. Just change base_url and your key.

from openai import OpenAI

client = OpenAI(
    api_key="tl-xxxxxxxxxxxxxxxxxxxxxxxx",
    base_url="https://api.thalam.ai/v1",
)

response = client.chat.completions.create(
    model="deepseek/deepseek-v4-flash-0731",
    messages=[
        {"role": "user", "content": "Hello from Thalam!"}
    ],
)

print(response.choices[0].message.content)