Model comparison
GPT-6 Sol vs GPT-6 Luna: when a twentieth of the price fits
GPT-6 Luna costs a twentieth of GPT-6 Sol per token. What OpenAI and the Codex docs say each tier is for, and what the gap means for coding sessions.
· Prices as of September 28, 2026
GPT-6 Sol
OpenAI · Released September 22, 2026
The mid-priced GPT-6 model, which OpenAI pitches for complex coding and agent workflows and which the Codex docs recommend for complex coding.
GPT-6 Sol facts and comparisonsGPT-6 Luna
OpenAI · Released September 22, 2026
The cheapest GPT-6 model, pitched for focused, high-volume, repeatable work, including narrower coding tasks.
GPT-6 Luna facts and comparisons
The short answer
GPT-6 Sol costs 20x as much as GPT-6 Luna per token, so a cache-heavy agentic coding session comes to $2.10 on Sol and $0.11 on Luna. The Codex docs recommend Sol for complex coding and Luna for focused, repeatable tasks. Running Sol on the hard problems and Luna on narrow, high-volume work is the split OpenAI's own positioning points to.
Choose GPT-6 Sol if
- Your work is complex coding and agentic workflows, which OpenAI names as the job GPT-6 Sol is built for.
- You follow the Codex docs, which name GPT-6 Sol for complex coding.
- You are replacing GPT-5.6 Sol, GPT-5.6 Terra, or GPT-5.4, which Codex suggests moving to Sol.
Choose GPT-6 Luna if
- You run focused, high-volume, repeatable tasks, the work OpenAI pitches GPT-6 Luna for.
- You are on a ChatGPT Free or Go plan, which gets Luna in the Codex app.
- You want to raise reasoning effort cheaply: Luna supports effort up to max in Codex, at $0.50 per million output tokens.
- You run many small jobs, where 110 sessions a month cost $11.55 rather than $231.00.
Side by side
Specs and prices
| Fact | GPT-6 Sol | GPT-6 Luna |
|---|---|---|
| Maker | OpenAI | OpenAI |
| API model id | gpt-6-sol | gpt-6-luna |
| Released | September 22, 2026 | September 22, 2026 |
| Status | Current | Current |
| Context window | 1.05M tokens | 1.05M tokens |
| Max output | 128K tokens | 128K tokens |
| Open weights | No | No |
| Input, per 1M tokens | $2 | $0.10 |
| Cache hit, per 1M | $0.20 | $0.01 |
| Cache write, per 1M | $2.50 | $0.125 |
| Output, per 1M tokens | $10 | $0.50 |
| Runs in | Codex, OpenCode, OpenRouter, and GitHub Copilot | Codex, OpenCode, OpenRouter, and GitHub Copilot |
Standard API rates in US dollars, as published by each maker on September 28, 2026. Batch and priority tiers, taxes, and subscription plans are not included. GPT-6 Sol: Requests over 272K input tokens cost 2x for input and cache and 1.5x for output, for the whole request. GPT-6 Luna: Requests over 272K input tokens cost 2x for input and cache and 1.5x for output, for the whole request.
Cost
What the same work costs
The same token counts, priced at each model’s published rates. Cached tokens are billed at each maker’s cache prices, so the session shows what caching is worth on each model.
Real sessions differ: the two models count the same code as different numbers of tokens, and reasoning settings change how much each one writes. Your own history is the real test.
| Workload | GPT-6 Sol | GPT-6 Luna |
|---|---|---|
| Agentic coding session, 100K input, 400K written to cache, 2M read from cache, 50K output | $2.10 | $0.11 |
| Large one-off review, 150K input with no cache hits, 10K output | $0.40 | $0.02 |
| Output-heavy generation, 30K input, 80K output | $0.86 | $0.04 |
| A month of sessions, 110 sessions: 5 a day, 22 working days | $231.00 | $11.55 |
| Where the session’s cost goes | ||
| Cache writes | $1.00 | $0.05 |
| Cache reads | $0.40 | $0.02 |
| Uncached input | $0.20 | $0.01 |
| Output | $0.50 | $0.03 |
- caching saves on the session with GPT-6 Sol (62%)
- $3.40
- caching saves on the session with GPT-6 Luna (61%)
- $0.17
How big is the price gap between GPT-6 Sol and GPT-6 Luna?
Twenty to one, on every rate. GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, with cache writes at $2.50 and cache hits at $0.20. GPT-6 Luna costs $0.10 and $0.50, with cache writes at $0.125 and hits at $0.01. Both follow OpenAI's rules from GPT-5.6 on: a write costs 1.25x input and a hit 0.1x.
On the example workloads the gap reads 19.1x for the session, 20x for the one-off review, and 21.5x for the output-heavy generation. The spread comes from rounding, not from the rates. Luna's totals are small enough that rounding each to the cent moves the ratio: the session is $0.11 on Luna against $2.10 on Sol, and the generation $0.04 against $0.86.
Over a month of 110 sessions, rounding stops mattering. Sol comes to $231.00 and Luna to $11.55, a difference of $219.45 at published API rates. At Luna's prices a single developer's choice rarely turns on money. It turns on whether the model handles the task.
What OpenAI says each GPT-6 tier is for
OpenAI describes Sol as built to power complex coding and agentic workflows, and the Codex docs recommend it for complex coding. It calls Luna its most efficient model for focused, high-volume tasks, including narrower coding tasks, and the Codex docs recommend it for focused, repeatable work. The two descriptions divide work by its shape: open-ended coding for Sol, narrow and repeatable jobs for Luna.
OpenAI adds two claims based on its factuality evaluation. It says Sol makes about half as many mistakes as its predecessor, and that Luna, at higher effort, matches GPT-5.6 Sol at about a hundredth of the cost. Both are the maker's claims, and both measure factual accuracy rather than coding.
Both models default to medium reasoning effort, and in Codex, Luna supports effort up to max. Reasoning tokens are billed as output, so Luna at a high effort setting still costs less than Sol at medium unless it writes 20x the tokens.
Running GPT-6 Sol and GPT-6 Luna side by side
Both run in Codex, OpenCode, OpenRouter, and GitHub Copilot, and both have a 1.05M context window with 128K of output. Luna is not available in Codex cloud. Neither is the default in Codex CLI, whose bundled model list starts on GPT-6 Astra.
Cache writes are the largest part of the session on both models: $1.00 of Sol's $2.10, or 48%, and $0.05 of Luna's $0.11, or 45%. Caching saves $3.40 on Sol and $0.17 on Luna compared with sending the same tokens uncached. On both, a request above 272K input tokens is billed at 2x for input and cache and 1.5x for output.
If you split work between the two tiers, the useful number is how your spending divides between them. EveryToken prices each local Codex and OpenCode request at OpenAI's API rates, per model, and shows what caching saved.
Prompt caching
How OpenAI bills cached tokens
OpenAI
Prompt caching is on by default. From GPT-5.6 on you can also mark up to four explicit cache breakpoints, while GPT-5.5 and earlier cache automatically only.
From GPT-5.6 on, a cache write costs 1.25x the uncached input price and a cache hit costs 0.1x. GPT-5.5 and earlier add no charge for writing the cache: written tokens are billed as ordinary input, and a hit costs 0.1x on the models compared here.
On GPT-5.6 and later, a cached prefix stays reusable for at least 30 minutes after its last use, and caching starts at 1,024 input tokens.
Source: OpenAI: Prompt caching
Your own numbers
See what GPT-6 Sol and GPT-6 Luna really cost you.
everyaitoken reads your Codex, OpenCode, and OpenRouter history on your Mac and prices every request at API rates, with what caching saved or cost. $9 once.
FAQ
Questions
What coding work does OpenAI recommend GPT-6 Luna for?
The Codex docs recommend GPT-6 Luna for focused, repeatable tasks, and OpenAI pitches it for focused, high-volume work, including narrower coding tasks. For complex coding, the Codex docs recommend GPT-6 Sol.
How much does a month of coding sessions cost on each?
At 110 example sessions a month, GPT-6 Sol comes to $231.00 and GPT-6 Luna to $11.55. These are API-equivalent estimates at published rates. Codex use through a ChatGPT plan is metered by the plan instead.
Do GPT-6 Sol and GPT-6 Luna cache the same way?
Yes. Both bill a cache write at 1.25x the input price and a cache hit at 0.1x. Cached prefixes remain reusable for at least 30 minutes after last use, and a prompt needs 1,024 input tokens before caching starts.
Were GPT-6 Sol and GPT-6 Luna released together?
Yes, both on September 22, 2026. GPT-6 Astra, the top tier, came out earlier, on September 3.