GPT-6 Astra
Newest frontier model for the most demanding agentic work. Highest intelligence and highest listed text credit rate.
Use for: exceptional complexity, major technical strategy, deep multi-step work.
Use the right intelligence for the job
Better work.
Fewer credits.
ChatGPT Business includes usage. Credits let eligible work continue after included allowances are reached. Model choice, task size, context, tools, and output length all affect consumption.
Rates verified against official OpenAI documentation · September 4, 2026
Start with the work,
then pick the model.
Select the closest task. Use the recommended model as your default, then move up only when complexity or consequences justify it.
Recommended default
Terra
Balanced credit use
Everyday professional work
Use Terra for reports, document analysis, drafting, coding, and work that needs sound judgment. It is the practical default for most substantive business tasks.
Move to Sol when: the problem is ambiguous, cross-functional, unusually difficult, or the cost of a wrong answer is high.
The current model family
In ChatGPT Work and Codex, the 5.6 family gives you a deliberate quality-to-throughput spectrum. Astra availability may depend on workspace rollout.
GPT-5.6 Sol
Built for hard reasoning, ambiguous problems, advanced coding, and high-stakes decisions.
Use for: security architecture, complex diagnosis, critical review, difficult implementation.
GPT-5.6 Terra
The balanced workhorse for production tasks, reporting, document analysis, coding, and sound judgment.
Use for: first drafts, analysis, plans, business writing, routine coding.
GPT-5.6 Luna
Fast and efficient for focused tasks where throughput matters more than deep deliberation.
Use for: extraction, classification, routing, summaries, structured cleanup.
How credits work
Rates are token-based, but the total for a task also reflects how much context the model reads, produces, retrieves, and processes with tools.
Included usage comes first
Business seats include an allowance for ChatGPT Work and Codex. If that allowance is exhausted, eligible usage can continue from purchased workspace credits when your administrator permits it.
1 · Included allowance
Used during normal Work and Codex activity
Used during normal Work and Codex activity
2 · Workspace credits
Used after the allowance, when enabled
Used after the allowance, when enabled
3 · Reset or switch
Wait for reset or use a smaller model if credits are unavailable
Wait for reset or use a smaller model if credits are unavailable
Estimate token-based credits
Enter token counts—not words. This estimates the published token-rate component and is not a billing guarantee.
Estimated usage2.00credits
At published GPT-5.6 Sol token rates.
Credit rate card
Credits per 1 million tokens. Cached input is substantially cheaper, so repeated context may cost less when caching applies.
| Model | Input | Cached input | Output | Relative output cost |
|---|---|---|---|---|
| GPT-6 Astra | 250 | 25 | 1,250 | Very high |
| GPT-5.6 Sol | 100 | 10 | 500 | High |
| GPT-5.6 Terra | 50 | 5 | 300 | Balanced |
| GPT-5.6 Luna | 5 | 0.5 | 30 | Lowest |
| GPT-5.5 earlier | 125 | 12.5 | 750 | Very high |
| GPT-5.4 earlier | 62.5 | 6.25 | 375 | Medium-high |
| GPT-5.4 mini earlier | 18.75 | 1.875 | 113 | Low |
| GPT-Image-2 · image | 200 | 50 | 750 | Varies by image |
| GPT-Image-2 · text | 125 | 31.25 | 250 | Varies by prompt |
Important: OpenAI reports that GPT-5.6 usage averages roughly 5–30 credits per message, but that is only an average. Fast mode increases consumption, and image generation typically uses included limits about 3–5× faster than similar turns without image generation.
Make the allowance last
A smaller, well-scoped task often beats an oversized prompt sent to the most powerful model.
Scope the source material
Attach only relevant files. Narrow repositories, date ranges, folders, and background context.
Specify the deliverable
Name the audience, format, depth, and length. Separate required work from optional polish.
Escalate intentionally
Start with Luna or Terra when appropriate. Move to Sol or Astra when the task—not habit—requires it.