Quick answer: choose GPT-6 Sol for difficult coding, research and multi-step professional work; choose GPT-6 Luna for high-volume, cost-sensitive tasks such as classification, extraction, routing and short summaries. OpenAI lists Sol at $2 per 1 million input tokens and $10 per 1 million output tokens, while Luna costs $0.10 input and $0.50 output. The API model IDs are gpt-6-sol and gpt-6-luna.[1]
30-second model selector
- Use Sol when a wrong answer, weak implementation or extra retry costs more than the model call.
- Use Luna when the task is narrow, repeatable and easy to verify automatically.
- Use both when Luna can process the routine majority and Sol can review difficult or low-confidence cases.
- Use Astra instead only when you need OpenAI’s highest-capability GPT-6 tier and the extra cost is justified.
GPT-6 Sol vs Luna pricing
OpenAI announced GPT-6 Sol and Luna on September 22, 2026 as faster, more affordable members of the GPT-6 family. Prices below are standard API token prices from the launch announcement and are quoted per 1 million tokens.[1]
| Model | Input / 1M tokens | Output / 1M tokens | Best starting use |
|---|---|---|---|
gpt-6-sol |
$2.00 | $10.00 | Complex coding, research, analysis and agents |
gpt-6-luna |
$0.10 | $0.50 | High-volume, bounded and easy-to-check tasks |
Real cost examples
The examples below apply the published input and output rates. They exclude any separate tool, storage, network or platform charges.
| Example usage | GPT-6 Sol | GPT-6 Luna |
|---|---|---|
| 5,000 input + 1,000 output tokens | $0.0200 | $0.0010 |
| 100,000 input + 20,000 output tokens | $0.4000 | $0.0200 |
| 1 million input + 200,000 output tokens | $4.0000 | $0.2000 |
estimated cost = (input tokens / 1,000,000 × input price) + (output tokens / 1,000,000 × output price)
Which model should you use?
Choose GPT-6 Sol for difficult work
Sol is the stronger default when the task needs deeper reasoning, careful coding, longer agent loops, tool use or professional-quality analysis. OpenAI reports that Sol improves over GPT-5.6 Sol across coding, factuality, computer use and professional-work evaluations while reducing API prices.[1]
- Repository-wide code changes and debugging
- Research that must reconcile several sources
- Complex document, spreadsheet or operational workflows
- Planning where a weak decision creates expensive downstream work
- Reviewing Luna outputs that fail a confidence or validation check
Choose GPT-6 Luna for scale
Luna is the cost-first choice for work that is well specified and can be checked with rules, schemas, tests or human sampling. OpenAI says Luna improves on its predecessor while targeting faster, more affordable everyday use.[1]
- Classification, tagging and intent detection
- Structured extraction from consistent documents
- Short summaries and content routing
- Draft variants that receive later review
- High-volume support triage with safe escalation
Use a two-model routing policy
A practical production pattern is to send bounded requests to Luna and escalate only hard or uncertain cases to Sol. Do not route solely by prompt length. Route by risk, complexity and how easily the result can be verified.
Use Luna when: - the output has a strict schema - an automated check can catch mistakes - the task is repetitive and low-risk Escalate to Sol when: - Luna returns low confidence or invalid output - the task needs multi-step reasoning or tool use - the result affects money, customers, production systems or reputation - a human reviewer flags ambiguity
How to access GPT-6 Sol and Luna
At launch, OpenAI said both models were available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users. Free and Go users can access GPT-6 Luna in the desktop app. OpenAI also stated that Sol and Luna were not yet available in regular Chat, and that the rollout in supported products would happen gradually. Availability can therefore differ by product, account and rollout stage.[1]
- Open the supported ChatGPT Work, Codex or desktop experience.
- Check the model selector for GPT-6 Sol or GPT-6 Luna.
- If a model is missing, verify your plan and try later; OpenAI described a gradual rollout.
- For API use, create or reuse a project API key and call the Responses API with the exact model ID.
GPT-6 API quickstart
The minimal request below uses OpenAI’s Responses API. Keep your key in an environment variable; never paste it into public code, browser-side JavaScript or a WordPress page.
export OPENAI_API_KEY="your_private_key"
curl https://api.openai.com/v1/responses \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-sol",
"input": "Review this migration plan. List data-loss risks, missing checks and a safer rollout order."
}'
To test Luna, change only the model value to gpt-6-luna. Before migrating production traffic, compare both models on a fixed evaluation set drawn from your real workload.
How to reduce GPT-6 API cost with caching
OpenAI says GPT-6 prompt caching can discount cached input-token reads by up to 90%. Eligible shared prefixes reused within a 30-minute window can benefit, and GPT-6 includes a caching dashboard, diagnostics, explicit cache breakpoints and prewarming controls.[2]
- Keep stable content first: place reusable instructions, tool definitions and reference material before changing user content.
- Do not reorder tools casually: OpenAI advises keeping tool definitions, schemas and ordering stable when possible.
- Append instruction changes: add later developer messages rather than rewriting the entire prefix.
- Measure cache hits: inspect the Prompt Caching Dashboard instead of assuming caching works.
- Diagnose misses: compare a request with a recent response to find changes that prevented reuse.
Migration checklist
- Create 20–50 representative tasks with expected results and pass/fail criteria.
- Record quality, latency, input tokens, output tokens, retries and total task cost.
- Test Sol and Luna with the same prompts and tools.
- Use structured outputs or deterministic checks where possible.
- Review tool permissions and require confirmation for consequential actions.
- Start with a small traffic percentage and keep a rollback route.
- Monitor caching separately from base token spend.
- Recheck official pricing and availability before making budget commitments.
Frequently asked questions
How much does GPT-6 Sol cost?
OpenAI lists GPT-6 Sol at $2 per 1 million input tokens and $10 per 1 million output tokens.[1]
How much does GPT-6 Luna cost?
OpenAI lists GPT-6 Luna at $0.10 per 1 million input tokens and $0.50 per 1 million output tokens.[1]
What are the GPT-6 Sol and Luna API model IDs?
The official model IDs are gpt-6-sol and gpt-6-luna.[1]
Is GPT-6 Luna available to free users?
OpenAI says Free and Go users can access GPT-6 Luna in the desktop app. That does not mean unrestricted access in every ChatGPT surface, and rollout visibility may vary.[1]
Is GPT-6 Sol better than Luna?
Sol is the better choice for difficult, quality-sensitive work; Luna is the better choice when cost and scale matter and the task is bounded and verifiable. “Better” depends on the workload, so test both on real examples.
Does prompt caching reduce GPT-6 costs?
Yes. OpenAI says cached input-token reads can receive discounts of up to 90%, provided the request has an eligible reusable prefix.[2]
Last checked: September 25, 2026. Pricing and availability can change.
Official sources