GPT-6 Sol vs Luna: Differences, Pricing and Which to Use

GPT-6 Sol and GPT-6 Luna are the two smaller members of OpenAI's GPT-6 family, released on 22 September 2026 under GPT-6 Astra. Sol is built for complex coding, agentic workflows and professional work; Luna is OpenAI's most efficient model for focused, high-volume jobs. Both have a 1,050,000-token context window, and on the API Sol costs $2 in and $10 out per million tokens while Luna costs $0.10 and $0.50 (as of October 2026). In ChatGPT, GPT-6 runs on Sol for paid plans and on Luna for Free and Go.

Key facts: GPT-6 Sol vs GPT-6 Luna

What GPT-6 Sol and Luna are

OpenAI's announcement frames the family as three tiers: Astra for the hardest work, Sol for everyday professional and coding work, and Luna for scale. OpenAI says it trained Sol and Luna with methods similar to Astra's and passed cost savings from caching and inference improvements on as a 50% price cut against the GPT-5.6 versions:

GPT-6 Sol: who it's for

Pick Sol for coding agents, multi-step business automations, document-heavy analysis and anything where a wrong answer costs more than the tokens. OpenAI's own figures (vendor-reported, at OpenAI's chosen effort settings):

• DeepSWE v1.1: 68.8% at max effort.
• AutomationBench 1.0.6: 33.2% at xhigh effort, at $0.27 per task.
• OSWorld 2.0 offline: 60.5% at xhigh effort.
• Agents' Last Exam: 56.4% at max effort.

OpenAI also says Sol makes about half as many factual mistakes as GPT-5.6 Sol on its internal factuality evaluation.

GPT-6 Luna: who it's for

Pick Luna for classification, extraction, routing, summarising support tickets, first-pass triage and any job you run thousands of times a day. OpenAI reports Luna at 66.6% on DeepSWE v1.1 at max effort, and says on its factuality evaluation Luna at higher effort matches GPT-5.6 Sol at about a hundredth of the cost. One caution from OpenAI's own GPT-6.1 Sol post: on a test of whether an agent admits its search tool is broken, Luna failed to disclose the problem in 28.7% of deliberately difficult cases, against 4.9% for GPT-6 Sol. If your agent relies on tools and must report failures honestly, test Luna on exactly that.

Sol vs Luna: how to choose

A practical pattern: route by task. Send simple, high-volume calls to Luna, escalate to Sol when a check fails or the task needs multiple tools, and keep Astra for the rare hard case. That's the same tiering idea behind most AI agent frameworks with model routing.

Availability

In ChatGPT

On 7 October 2026 OpenAI started rolling out "GPT-6 with Intelligent UI" in the Chat tab: answers can include interactive components such as charts, buttons, forms and small tools. OpenAI's availability note says GPT-6 in ChatGPT is powered by Sol on Plus, Pro, Business and Enterprise and by Luna on Free and Go, with Free and Go starting the following day (8 October). Enterprise availability depends on admin settings.

Before that, from 22 September, Sol and Luna were in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu, and Free and Go users could use Luna in the desktop app. OpenAI says the 7 October update doesn't change the models powering Work and Codex.

On the API

Use gpt-6-sol or gpt-6-luna through the Responses API, Chat Completions or Batch. Built-in tools (web search, file search, code interpreter, computer use, MCP and others) work through the Responses API; Chat Completions supports function calling only when reasoning effort is none.

from openai import OpenAI

client = OpenAI()

response = client.responses.create(
    model="gpt-6-luna",
    reasoning={"effort": "low"},
    input="Classify this ticket as billing, bug or feature request: ...",
)

print(response.output_text)

If you're new to tool calling, our function calling and MCP explainers cover the basics.

Pricing in detail (as of October 2026)

From OpenAI's pricing page, per 1M tokens, Standard tier:

Batch and Flex are half these rates; Fast mode (formerly Priority) is double. Regional processing adds 10%. Cache writes are billed at 1.25x the uncached input rate ($2.50 for Sol, $0.125 for Luna).

The long-prompt jump is easy to miss: cross 272K input tokens and the whole request is billed at the higher rate. If you're stuffing big contexts, context engineering usually saves more than switching models.

What about GPT-6.1 Sol?

A week after launch, on 29 September, OpenAI released gpt-6.1-sol, an upgrade it says "nearly matches GPT-6 Astra's intelligence" on agentic coding, computer use and professional work at one fifth of Astra's token prices. It keeps $2 input and $10 output, cuts cached input to $0.10, drops the none reasoning effort (levels are low to max), and has an Ultrafast mode at 6x the standard price. It's in ChatGPT Work and Codex for paid plans and on the API. If you're starting a new Sol integration today, test 6.1 Sol first.

Bottom line

Sol is the default for work that matters; Luna is the default for work that repeats. The 20x price gap is big enough that most production systems should use both. For how these compare with Google's newest model, see Gemini 4 Argon vs GPT-6.