GPT-6 Astra guide
Chat Completions, Responses API, Codex configuration, parameter compatibility, and timeout troubleshooting for GPT-6 Astra.
gpt-6-astra uses the same Models Hub base URL as GPT-5.6 and supports both Chat Completions and the Responses API. The main difference is stricter parameter compatibility, not a different URL. Prefer the Responses API for Codex, tool calling, and multi-turn reasoning.
APIs and parameters
| Use case | Endpoint | Output limit | Notes |
|---|---|---|---|
| Basic text chat | /v1/chat/completions | max_completion_tokens | Models Hub converts max_tokens from older clients |
| Codex, tool calling, multi-turn reasoning | /v1/responses | max_output_tokens | Recommended; Codex uses this protocol |
Do not copy every GPT-5.6 parameter
GPT-6 does not support temperature, top_p, or top_logprobs; do not send logprobs with Chat Completions either. GPT-6 does not support reasoning_effort = "none"; start with low for latency-sensitive workloads.
Responses API (recommended)
curl https://modelsok.com/v1/responses \
-H "Authorization: Bearer $MODELSOK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"input": "Introduce Models Hub in one sentence.",
"reasoning": {"effort": "low"},
"max_output_tokens": 256,
"stream": true
}'Use the Responses API for tool calling, Codex, and tasks that preserve reasoning context. See the full Responses reference.
Chat Completions
curl https://modelsok.com/v1/chat/completions \
-H "Authorization: Bearer $MODELSOK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"messages": [
{"role": "user", "content": "Introduce Models Hub in one sentence."}
],
"reasoning_effort": "low",
"max_completion_tokens": 256,
"stream": true
}'If an older SDK still sends max_tokens, Models Hub converts it to max_completion_tokens only for the GPT-6 model family. An explicitly supplied new field is preserved, and other models are unaffected. New integrations should send max_completion_tokens directly.
Codex configuration
See Use Codex CLI with Models Hub for installation and verification. The key configuration is:
model = "gpt-6-astra"
model_provider = "modelsok"
model_reasoning_effort = "low"
[model_providers.modelsok]
name = "Models Hub"
base_url = "https://modelsok.com/v1"
env_key = "MODELSOK_API_KEY"
wire_api = "responses"
stream_idle_timeout_ms = 300000524 or slow time to first token
Streaming improves perceived latency but cannot eliminate a 524 caused by an upstream service that sends no data for too long. Retry with reasoning effort set to low. If the issue persists, collect the Request ID, Upstream Request ID, channel number, time to first token, and total duration from Models Hub usage logs for support.