GPT-6 (Astra / Sol / Luna)

GPT-6 is an OpenAI family of three general-purpose models with identical capabilities in different price and speed tiers: astra — the flagship for the hardest tasks, sol — the balanced workhorse, luna — fast and cheap for everyday queries. All three accept images and files, reason with a controllable depth, and have native web search.

Parameters and capabilities

Parameter	Value
model	openai/gpt-6-astra, openai/gpt-6-sol, openai/gpt-6-luna (short aliases work too)
reasoning_effort	low, medium, high, xhigh, max
Disabling reasoning	Astra — no (always thinks); Sol and Luna — yes (off)
Images on input	yes (image_url, data URLs)
Files on input	yes (PDF/DOCX/TXT etc., up to 5)
Web search	native (web_search: "native")
Cache read	Astra $1.25 / Sol $0.25 / Luna $0.012 per 1M

Picking a tier

Model	Input/output per 1M	Profile
gpt-6-astra	$12.5 / $62.5	maximum quality: architecture, deep analysis, hard agents
gpt-6-sol	$2.5 / $12.5	price/quality balance for development and long sessions
gpt-6-luna	$0.125 / $0.625	bulk queries, classification, drafts

Example

curl https://api.mixen.ai/v1/chat/completions \
  -H "Authorization: Bearer $MIXEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-6-sol",
    "reasoning_effort": "high",
    "messages": [{"role": "user", "content": "Design a multi-tenant SaaS database schema and justify the choices"}]
  }'

Web search — via "web_search": "native"; sources arrive in message.annotations.

Strengths and limitations

• Same capabilities in three tiers — switch model to match the budget without rewriting code.
• Reasoning eats max_tokens: set a generous ceiling for long answers (finish_reason: "length" means truncation).
• Astra always thinks: reasoning tokens are billed on every request.
• Context cache fires automatically on a repeated prefix — keep the system prompt stable.

All models — in the catalog. The general workflow — chat guide.