OpenAI has released GPT-6 Sol for demanding professional work and GPT-6 Luna for high-volume tasks. Both publish 1.05-million-token context windows, but price, access and the long-context surcharge change the buying decision.
RELATED BUYER PROFILEReview GPT-6 Sol pricing, access and test-pending evidencePricing · access · strengths · limitations →RELATED DECISION GUIDERun a tiered same-work model pilotWorkflow · evidence · risk · governance →CONTINUE THE DECISIONReview the lower-cost GPT-6 Luna profileEvidence · workflow · next action →CONTINUE THE DECISIONAdd GPT-6 Sol to a current shortlistEvidence · workflow · next action →CONTINUE THE DECISIONMeasure complete cost per accepted taskEvidence · workflow · next action →CONTINUE THE DECISIONBuild a controlled developer model stackEvidence · workflow · next action →What you need to know
- The public API model IDs are gpt-6-sol and gpt-6-luna; both list a 1,050,000-token context window and 128,000-token maximum output
- Standard API rates are $2 input and $10 output per million tokens for Sol, versus $0.10 input and $0.50 output for Luna
- Requests above 272,000 input tokens price the full request at twice the input/cache rate and 1.5 times the output rate, so a one-million-token window is not a flat-cost entitlement
The verified buying boundary
Short answer: two real releases, not a benchmark verdict
OpenAI announced GPT-6 Sol and GPT-6 Luna on 22 September 2026. Sol is positioned for demanding coding and professional work; Luna is positioned for faster, high-volume tasks. Both are available through the API and are rolling into ChatGPT Work, Codex and eligible GitHub Copilot plans. The model identifiers, published rates and product routes are verified. Performance, factuality and alignment comparisons in the launch material remain provider-reported until independently reproduced.
Start with Luna; move work to Sol only with evidence
Luna's standard token rate is one twentieth of Sol's before other charges. That makes it the sensible baseline for classification, extraction, formatting, routing, first-pass support and other high-volume work. Sol is the stronger candidate for complex repository work, longer agent plans and professional tasks where a more capable model can reduce failed attempts or review. A routing policy should be based on accepted-task evidence, not model names.
Published API prices are clear—but incomplete
Sol lists $2 per million input tokens, $0.20 cached input, $2.50 cache write and $10 output. Luna lists $0.10 input, $0.01 cached input, $0.125 cache write and $0.50 output. Batch and Flex processing are listed at 50% of standard rates, while Fast processing is twice the standard rate. Regional processing adds 10%, and EU data residency is listed only for the standard service tier. Tools, retries, infrastructure and human review remain additional workflow costs.
The long-context surcharge changes the one-million-token story
Both model pages publish a 1,050,000-token context window and a 128,000-token output limit. OpenAI's pricing documentation states that when input exceeds 272,000 tokens, the full request is charged at twice the input and cache rate and 1.5 times the output rate. Buyers should test retrieval, compaction and caching before sending an entire repository or document estate on every turn.
Access depends on the surface
OpenAI says Sol and Luna are available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu accounts, while Free and Go users receive Luna in the desktop app. The launch also says the models were not yet available in the conversational Chat surface at announcement time, and rollout is gradual. GitHub separately lists Sol for Pro+, Max, Business and Enterprise, and Luna for Pro, Pro+, Max, Business and Enterprise, with usage-based billing and administrator model policy.
Shared specifications do not mean interchangeable behaviour
Both API models accept text and image input, produce text output and expose reasoning levels from none through max. Sol lists an April 2026 knowledge cutoff and Luna a May 2026 cutoff. Those facts do not establish which one follows a company's instructions, uses tools reliably or produces an acceptable result on a particular task. Test the exact API or product surface the team will buy.
Run a tiered same-work pilot
Freeze ten representative tasks, inputs, tools, permissions and acceptance tests. Run Luna first, then Sol only on the failures and high-value cases. Record accepted-task rate, unsupported claims or changes, tests passed, tool failures, retries, cache hits, latency, reviewer minutes and billed cost. Add a routing rule only where the more expensive route produces a repeatable improvement after review.
HubAI buyer verdict
The family creates a useful price ladder: Luna is the default candidate for volume, while Sol belongs on a shortlist for harder coding and agent work. Neither receives a HubAI Score before an independent same-task evaluation. Buyers should begin below 272K input, use standard processing and compare complete cost per accepted result before enabling long contexts, Fast processing or wider agent permissions.
Evidence limits and editorial disclosure
Independent editorial coverage; not sponsored. Pricing, model specifications, access, benchmark and safety statements are attributed to OpenAI, GitHub and their official documentation, checked on 23 September 2026. HubAI has not independently tested either model. The visible UK Google Trends first 25 contained no direct GPT-6 Sol or Luna query during this review, so no search-volume or breakout claim is made. The cover is an original conceptual illustration, not a product interface, provider logo or performance chart.
HUBAI VIEWUse Luna as the low-cost baseline and promote work to Sol only when a frozen same-task test shows a measurable gain in accepted output, review time or failure recovery.
Buyer decision signal: New model family · API, ChatGPT Work and Copilot rollout
What to verify next
1Confirm the intended route: OpenAI API, ChatGPT Work, Codex or GitHub Copilot
2Start with Luna on frozen representative tasks and written acceptance tests
3Escalate only failed or high-value tasks to Sol and record the reason
4Keep initial inputs below 272K and measure cache behaviour before expanding context
5Record tokens, tools, retries, infrastructure and human review per accepted result
6Verify account access, administrator policy, regional processing and data controls
7Keep human approval, logs, rollback and an alternative model for consequential workflows
Read the evidence
Capabilities, availability and prices can change. HubAI keeps analysis separate from the underlying official material.
01OpenAI: Introducing GPT-6 Sol and GPT-6 Luna — 22 September 2026Open source ↗02OpenAI developer documentation: GPT-6 SolOpen source ↗03OpenAI developer documentation: GPT-6 LunaOpen source ↗04OpenAI API pricing and long-context ratesOpen source ↗05GitHub: GPT-6 Sol and Luna rollout in Copilot — 22 September 2026Open source ↗
Models
Models
Models
Models