Day 0 Support: GPT-6 Astra

LiteLLM now supports gpt-6-astra, OpenAI's next major model. Route traffic to it through the LiteLLM AI Gateway with the same config you use for every other OpenAI model.

Astra is the first model on the GPT-6 name, and OpenAI's own research post previewed it producing machine-checked proofs for ten open problems in mathematics and theoretical computer science. Through the API it behaves like the GPT-5 reasoning line: max_completion_tokens instead of max_tokens, reasoning_effort up to xhigh, no temperature while reasoning is on, prompt caching, and the long-context pricing tier above 272K input tokens.

note

Cost tracking works on the version you already run. Hit the Reload Model Cost Map button in the Admin UI (or POST /reload/model_cost_map) to pull the gpt-6-astra pricing from GitHub. This feature is available on v1.76.0 and above.

Parameter handling needs the next release. The GPT-5 reasoning classifier in LiteLLM matched gpt-5* names only, so on older versions a gpt-6-astra request keeps max_tokens and temperature as sent and OpenAI rejects them. The fix widening it to GPT-6 is on main now and ships in this Saturday's release candidate; until you upgrade, send max_completion_tokens yourself and leave temperature unset.

Usage

LiteLLM Proxy

1. Setup config.yaml

model_list:
  - model_name: gpt-6-astra
    litellm_params:
      model: openai/gpt-6-astra
      api_key: os.environ/OPENAI_API_KEY

2. Start the proxy

docker run -d \
  -p 4000:4000 \
  -e OPENAI_API_KEY=$OPENAI_API_KEY \
  -v $(pwd)/config.yaml:/app/config.yaml \
  ghcr.io/berriai/litellm:main-latest \
  --config /app/config.yaml

3. Test it

curl -X POST "http://0.0.0.0:4000/chat/completions" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $LITELLM_KEY" \
  -d '{
    "model": "gpt-6-astra",
    "messages": [
      {"role": "user", "content": "Prove that there are infinitely many primes."}
    ],
    "reasoning_effort": "high"
  }'

LiteLLM Python SDK

from litellm import completion

response = completion(
    model="openai/gpt-6-astra",
    messages=[
        {"role": "user", "content": "Prove that there are infinitely many primes."}
    ],
    reasoning_effort="high",
)

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

Responses API

For agentic and multi-turn workflows, use /v1/responses to preserve reasoning state across turns. A litellm.completion() call that combines function tools with active reasoning is bridged to /v1/responses automatically, the same way it is for GPT-5.4 and newer.

curl -X POST "http://0.0.0.0:4000/v1/responses" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $LITELLM_KEY" \
  -d '{
    "model": "gpt-6-astra",
    "input": "Plan and write a Python script that scrapes a webpage and summarizes it.",
    "reasoning": {"effort": "high"}
  }'

Pricing

Prices are per 1M tokens (USD), shown as short context (≤272K tokens) / long context (>272K tokens).

Model	Input	Cached input	Cache write	Output
gpt-6-astra	$10.00 / $20.00	$1.00 / $2.00	$12.50 / $25.00	$50.00 / $75.00

The flex service tier is billed at half the standard rate and priority at double, on both the short and long context tiers, and the Batch API is billed at half the standard input and output rate. Pass service_tier on the request and LiteLLM picks the matching rate.

Notes

• gpt-6-astra supports reasoning_effort values none, low, medium, high, and xhigh; minimal is not accepted.
• Availability is rolling out through the API; check your OpenAI account for model access.
• See the OpenAI provider docs for the full parameter reference.