Skip to content

Model comparison

GPT-6 Astra vs GPT-6 Sol: Codex default or a fifth the cost?

GPT-6 Astra costs 5x as much as GPT-6 Sol on every token, cached or not. What OpenAI built each for, which one Codex picks, and what a session costs.

· Prices as of September 28, 2026

  • GPT-6 Astra

    OpenAI · Released September 3, 2026

    OpenAI's top GPT-6 model and its most expensive, aimed at the hardest long-running work that spans many tools, including coding.

    GPT-6 Astra facts and comparisons

  • 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 comparisons

The short answer

GPT-6 Astra charges exactly 5x the GPT-6 Sol rate on input, output, cache writes, and cache hits, so the example agentic coding session costs $10.50 on Astra and $2.10 on Sol. OpenAI pitches Astra for the hardest end-to-end work and Sol for complex coding and agent workflows, and the Codex docs recommend Sol for complex coding. Astra is still the default in Codex CLI's bundled model list, so check which one your sessions actually run on.

Choose GPT-6 Astra if

  • You take on the hardest long-running jobs that span many tools, which OpenAI names as the work GPT-6 Astra is built for.
  • You keep Codex CLI's default model, which is Astra in the bundled model list of version 0.158.0, starting at low reasoning effort.
  • Your tasks run long, and you want the model OpenAI says stays coherent during long tasks better than GPT-5.6 Sol and earlier models.
  • You judge cost per finished task rather than per token, since OpenAI says Astra reached stronger results with substantially fewer output tokens in several evaluations.

Choose GPT-6 Sol if

  • You want the model the Codex docs recommend for complex coding.
  • You pay API rates for many sessions: 110 a month come to $231.00 on Sol against $1,155.00 on Astra.
  • You prefer a medium reasoning effort by default, which Sol uses in both the API and Codex.
  • You are moving off GPT-5.6 Sol, GPT-5.6 Terra, or GPT-5.4, the models Codex suggests replacing with GPT-6 Sol.

Side by side

Specs and prices

FactGPT-6 AstraGPT-6 Sol
MakerOpenAIOpenAI
API model idgpt-6-astragpt-6-sol
ReleasedSeptember 3, 2026September 22, 2026
StatusCurrentCurrent
Context window1.05M tokens1.05M tokens
Max output128K tokens128K tokens
Open weightsNoNo
Input, per 1M tokens$10$2
Cache hit, per 1M$1$0.20
Cache write, per 1M$12.50$2.50
Output, per 1M tokens$50$10
Runs inCodex, OpenCode, OpenRouter, and GitHub CopilotCodex, 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 Astra: Requests over 272K input tokens cost 2x for input and cache and 1.5x for output, for the whole request. GPT-6 Sol: 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.

Example workload costs
WorkloadGPT-6 AstraGPT-6 Sol
Agentic coding session, 100K input, 400K written to cache, 2M read from cache, 50K output$10.50$2.10
Large one-off review, 150K input with no cache hits, 10K output$2.00$0.40
Output-heavy generation, 30K input, 80K output$4.30$0.86
A month of sessions, 110 sessions: 5 a day, 22 working days$1,155.00$231.00
Where the session’s cost goes
Cache writes$5.00$1.00
Cache reads$2.00$0.40
Uncached input$1.00$0.20
Output$2.50$0.50
caching saves on the session with GPT-6 Astra (62%)
$17.00
caching saves on the session with GPT-6 Sol (62%)
$3.40

Is GPT-6 Astra really 5x the price of GPT-6 Sol?

Yes, on every line of the price sheet. GPT-6 Astra charges $10 per million input tokens and $50 per million output tokens, while GPT-6 Sol charges $2 and $10. Cache writes cost $12.50 against $2.50, and cache hits cost $1 against $0.20. Every ratio is 5x.

Because the ratio never changes, caching does not narrow the gap here, the way a cheaper cache hit on one model can elsewhere. The example session costs $10.50 on Astra and $2.10 on Sol, a difference of $8.40. The uncached one-off review costs $2.00 against $0.40, and the output-heavy generation $4.30 against $0.86. All three land on 5x.

The session splits the same way on both models. Cache writes take 48%, output 24%, cache reads 19%, and fresh input 10%. So what moves spending on Astra is what moves it on Sol: how much context the tool writes to the cache, and how many tokens the model writes back.

Which GPT-6 model does Codex use by default?

Codex CLI's bundled model list, as of version 0.158.0, makes GPT-6 Astra the default and starts it at low reasoning effort. GPT-6 Sol defaults to medium effort in both the API and Codex. A default Codex setup therefore runs the pricier model, at a lighter effort setting than Sol would use.

The Codex docs lean the other way when they recommend a model. They name GPT-6 Sol for complex coding, and Codex suggests moving to Sol from GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.4. Astra runs in the Codex app, CLI, and IDE extension, but not in Codex cloud.

With ChatGPT sign-in, your plan's usage limits decide what a Codex session costs you, not these per-token rates. The dollar figures on this page are API-equivalent estimates at OpenAI's published rates. They matter most to API-key users, and they still show how far apart the two models sit.

Can fewer output tokens make GPT-6 Astra cheaper per task?

OpenAI says GPT-6 Astra reached stronger results with substantially fewer output tokens in several evaluations, for a lower estimated cost per task. That claim is about cost per task, and it does not name GPT-6 Sol as the comparison. At 5x the output rate, Astra would have to write about a fifth as many tokens as Sol just to match it on output cost.

Reasoning effort shifts the picture as well, since reasoning tokens are billed as output. Astra's low starting effort in Codex and Sol's medium default mean the two do not start from the same place. Compare them at the effort level you actually run, on your own tasks.

Both models share a 1.05M context window, accept up to 922K input tokens, and write up to 128K tokens of output. On both, a request over 272K input tokens costs 2x for input and cache and 1.5x for output, and the example session stays below that line. EveryToken reads your local Codex history and prices each request at these rates, so you can see which model each session used and 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 Astra and GPT-6 Sol 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.

Launching soonSee the cache math

FAQ

Questions

How much more does GPT-6 Astra cost than GPT-6 Sol?

Five times as much per token on input, output, cache writes, and cache hits. A month of 110 example sessions comes to $1,155.00 on Astra and $231.00 on Sol, a difference of $924.00 at published API rates.

Is GPT-6 Astra available in Codex cloud?

No. OpenAI offers Astra in the Codex app, CLI, and IDE extension, but not in Codex cloud. Codex cloud chats on ChatGPT plans run GPT-5.6 Sol.

Does caching save more on GPT-6 Astra?

In dollars, yes, because its input is dearer. Caching saves $17.00 on Astra and $3.40 on Sol in the example session, and both work out to 62% of the uncached cost. Both bill cache writes at 1.25x input and hits at 0.1x.

Which came out first, GPT-6 Astra or GPT-6 Sol?

GPT-6 Astra, on September 3, 2026. GPT-6 Sol and GPT-6 Luna followed on September 22, 2026, with the same context window and output limit.

Sources

  1. OpenAI: API pricing
  2. OpenAI docs: GPT-6 Astra
  3. OpenAI: Using the latest model
  4. OpenAI: API changelog
  5. Codex docs: Models
  6. Codex CLI: bundled model catalog
  7. OpenRouter: GPT-6 Astra
  8. OpenCode docs: Zen
  9. GitHub Docs: Supported AI models in Copilot
  10. OpenAI docs: GPT-6 Sol
  11. OpenAI: Introducing GPT-6 Sol and Luna
  12. OpenRouter: GPT-6 Sol
  13. OpenAI: Prompt caching

Related comparisons

  • GPT-6 Astra vs GPT-5.5

    GPT-6 Astra doubles GPT-5.5's input price and adds a 1.25x cache-write charge GPT-5.5 never had. What that does to agentic sessions and to Codex.

  • GPT-6 Astra vs GPT-6 Luna

    GPT-6 Astra and GPT-6 Luna sit at opposite ends of OpenAI's lineup, 100x apart per token. What each is for, and where GPT-6 Sol fits between them.

  • GPT-6 Sol vs GPT-5.6 Sol

    GPT-6 Sol costs half as much as GPT-5.6 Sol on every rate, and Codex suggests the move. What changes, what the promo pricing means, and what stays put.

  • GPT-6 Sol vs GPT-5.6 Terra

    GPT-6 Sol matches GPT-5.6 Terra's input and cache prices and charges less for output. What that means for coding sessions and the move Codex suggests.

  • GPT-6 Sol vs GPT-6 Luna

    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.

  • Claude Fable 5.1 vs GPT-6 Astra

    Claude Fable 5.1 and GPT-6 Astra share $10 input and $50 output prices, and a cached coding session costs $10.50 on each. Caching decides which way it tips.