Skip to content
API Reference

GPT

GPT models cover a broad capability range, from fast general-purpose chat to dedicated coding variants, with reasoning_effort control on the GPT-5 line.

Models

Model Code Notes
GPT 6.1 Sol gpt-6.1-sol GPT-6.1 line.
GPT 6 Astra gpt-6-astra Most capable GPT-6 reasoning model.
GPT 6 Sol gpt-6-sol GPT-6 line; balanced speed and cost for coding and agentic work.
GPT 6 Luna gpt-6-luna GPT-6 line; most efficient, for high-volume workloads.
GPT 5.6 Terra gpt-5.6-terra GPT-5.6 line.
GPT 5.6 Sol gpt-5.6-sol GPT-5.6 line.
GPT 5.6 Luna gpt-5.6-luna GPT-5.6 line.
GPT 5.5 gpt-5.5 GPT-5.5 line.
GPT 5.4 gpt-5.4 GPT-5.4 line.
GPT 5.4 Mini gpt-5.4-mini GPT-5.4 line, smaller.
GPT 5.4 Nano gpt-5.4-nano GPT-5.4 line, smallest/fastest.
GPT 5.4 Pro gpt-5.4-pro GPT-5.4 line, Responses-native. Chat Completions also works — the router bridges it to the Responses API.
GPT 5.3 Codex gpt-5.3-codex Agentic coding variant.
GPT 5.2 Codex gpt-5.2-codex Agentic coding variant.
GPT 5.1 gpt-5.1 GPT-5.1 line.
GPT 5.1 Codex gpt-5.1-codex Agentic coding variant.
GPT 5.1 Codex Max gpt-5.1-codex-max Agentic coding, highest capability.
GPT 5.1 Codex Mini gpt-5.1-codex-mini Agentic coding, cost-effective.
GPT 5 gpt-5 GPT-5 base.
GPT 5 Nano gpt-5-nano GPT-5 base, smallest/fastest.
GPT 4.1 gpt-4.1 Previous generation.
GPT 4.1 Mini gpt-4.1-mini Previous generation, smaller.

Capabilities

  • tools, tool_choice, parallel_tool_calls, functions (legacy)
  • reasoning_effort on all GPT-5 variants
  • logprobs / top_logprobs, logit_bias, web_search_options — with limits on GPT-5 (see the note below)
  • modalities, audio, prediction, service_tier

GPT-5 parameter sanitisation

GPT-5 models convert max_tokens → max_completion_tokens, accept temperature only at 1.0 (or with reasoning_effort = "none" on GPT-5.1, 5.2, 5.4 and 5.5), and drop stop, penalties and logit_bias. logprobs / top_logprobs also need reasoning_effort = "none" on those models. See Per-Model Details → GPT-5 series.

Request tips

  • GPT-5 and Codex models need max_tokens of at least 16.
  • Include the word "JSON" in the prompt when you use response_format: {"type": "json_object"}.
  • On /v1/chat/completions, GPT-5.6 and later (including GPT-6) reject function tools unless reasoning_effort is "none" (400 otherwise). Set it to "none", or use /v1/responses.

Example

from openai import OpenAI

client = OpenAI(base_url="https://llm.siraya.ai/v1", api_key="<API_KEY>")
resp = client.chat.completions.create(
    model="gpt-5.4",
    messages=[{"role": "user", "content": "Write a haiku about routing."}],
    reasoning_effort="medium",
)
print(resp.choices[0].message.content)