Back to changelog

GPT-6 Sol & Luna at Half the Price

OpenAI's GPT-6 Sol and GPT-6 Luna are live on the gateway at $2 / $10 and $0.10 / $0.50 per 1M tokens — Sol at half the GPT-5.6 Sol rate, Luna at half the GPT-5.6 Luna input rate and 58% below it on output — with cache reads at 10% of input, a 1.05M-token context window, and reasoning from none to max.

A glowing golden sun and a silver crescent moon on a circuit-board chip, surrounded by a halved coin stack with a price tag, a terminal window, a down arrow, and chat bubbles

The GPT-6 family launched with Astra at $10 / $50 per 1M tokens, a price that keeps it out of most high-volume pipelines. OpenAI's two new siblings close that gap: GPT-6 Sol and GPT-6 Luna are live on LLM Gateway today. Sol costs half of GPT-5.6 Sol on both input and output; Luna costs half of GPT-5.6 Luna on input and 58% less on output. Both keep the same 1.05M-token context window, 128K output ceiling, and pro reasoning mode.

Pricing

Model Input / 1M Cached input / 1M Cache write / 1M Output / 1M
gpt-6-sol $2.00 $0.20 $2.50 $10.00
gpt-6-luna $0.10 $0.01 $0.125 $0.50

Everything derived from the base rate follows the GPT-5.6 rules:

  • Cache reads bill at 10% of input and cache writes at 1.25x, with a single 30-minute TTL. Explicit caching via prompt_cache_options and prompt_cache_breakpoint is forwarded on both models.
  • Long context: a prompt above 272K input tokens bills the whole request at 2x on the input side and 1.5x on output — Sol at $4 / $15, Luna at $0.20 / $0.75.
  • Flex halves the bill (Sol $1 / $5, Luna $0.05 / $0.25) and Priority doubles it (Sol $4 / $20, Luna $0.20 / $1.00). Pass service_tier to pick one.
  • Web search costs $0.01 per call plus the search content tokens at model rates.

How they compare

Model Input / 1M Cached input / 1M Output / 1M
gpt-6-astra $10.00 $1.00 $50.00
gpt-5.6-sol (promo through Nov 21) $4.00 $0.40 $20.00
gpt-6-sol $2.00 $0.20 $10.00
gpt-5.6-luna $0.20 $0.02 $1.20
gpt-6-luna $0.10 $0.01 $0.50

Sol costs half of GPT-5.6 Sol's promotional rate and a fifth of GPT-6 Astra. Luna is half of GPT-5.6 Luna on input and 58% cheaper on output, a 1M-context reasoning model at GPT-4.1 nano input rates.

For a session that reads 1M input tokens, 800K of them from cache, and writes 100K output tokens:

Model Session cost
gpt-6-astra $7.80
gpt-5.6-sol $3.12
gpt-6-sol $1.56
gpt-5.6-luna $0.176
gpt-6-luna $0.078

What else changed

  • Reasoning accepts none, low, medium (the default), high, xhigh, and max, and both models take reasoning.mode: "pro" on top of the effort.
  • Knowledge cutoff moves to April 20, 2026 on Sol and May 18, 2026 on Luna.
  • Tool calls with reasoning on work from /v1/chat/completions. OpenAI's own Chat Completions endpoint only allows function calling on these models with reasoning off; the gateway forwards them to the Responses API, so nothing changes in your request.
  • The models accept max_tokens, verbosity, tools, tool_choice, and response_format. Sampling parameters such as temperature are stripped before forwarding, so a request that sets them still succeeds.
1curl https://api.llmgateway.io/v1/chat/completions \2    -H "Authorization: Bearer $LLM_GATEWAY_API_KEY" \3    -H "Content-Type: application/json" \4    -d '{5        "model": "openai/gpt-6-sol",6        "reasoning_effort": "high",7        "messages": [{ "role": "user", "content": "Refactor this module to remove the circular import." }]8    }'

Swap in openai/gpt-6-luna for classification, extraction, and other high-volume work. Both models route through OpenAI directly for now; Azure and Bedrock mappings follow once we have verified them end to end.

Service tiers docs → | GPT-6 Sol on the models page →