GPT-6 Sol and GPT-6 Luna are OpenAI's mid-tier and low-cost GPT-6 models, sitting below GPT-6 Astra. OpenAI says it trained them with methods similar to Astra's to bring its gains to faster, cheaper models. In the API they are gpt-6-sol and gpt-6-luna. Sol costs 2 US dollars per million input tokens and 10 per million output; Luna costs 0.10 and 0.50; Astra remains 10 and 50. OpenAI says the Sol and Luna prices are 50 percent below the GPT-5.6 Sol and Luna promotional prices. All three have a 1,050,000-token context window and 128,000 max output tokens. As a starting rule: use Luna for classification, extraction, routing, and other high-volume work; use Sol as the default for coding agents and everyday professional tasks; reserve Astra for the hardest tasks, and only after your own tests show Sol falling short. OpenAI's benchmark comparisons are against Claude Opus 5 and Fable, not against Claude Opus 5.5, which launched the same day.
What are GPT-6 Sol and GPT-6 Luna, and which GPT-6 model should you use?
GPT-6 Sol and GPT-6 Luna are the two lower-cost models OpenAI added to its GPT-6 family below GPT-6 Astra. Sol is priced at 2 US dollars per million input tokens and 10 per million output, Luna at 0.10 and 0.50, both half the promotional price of their GPT-5.6 predecessors. Both carry the same 1,050,000-token context window as Astra. Astra stays the top tier at 10 and 50. The practical choice is Luna for high-volume and simple work, Sol for most coding and agent work, and Astra only where a measured quality gap justifies five times the price of Sol.
Published · Updated · Evidence-linked, not search-volume ranked.
Why this question is current
Exact query-volume data was unavailable, so RepoRadar uses these as current demand and intent signals rather than a claimed volume ranking.
- stories with more than 60 points, trailing 48 hours · Hacker News Algolia search_by_date · global English-language developer community · checked 2026-09-23T22:26:00Z
The launch thread objectID 49805509 titled GPT-6 Sol and Luna, created 2026-09-22T18:00Z, carried 1728 points and 820 comments at check time, the second-highest story in the 48-hour window. Interest signal, not search volume. - gpt-6 sol / gpt 6 sol vs · Google Suggest · US · checked 2026-09-23T22:28:00Z
The seed gpt-6 sol completed to phrasings including gpt 6 sol terra luna, gpt 6 sol release date, gpt 6 sol vs opus 5, and gpt 6 sol vs astra. The seed gpt 6 sol vs completed to gpt 6 sol vs astra, gpt 6 sol vs terra vs luna, and gpt 6 sol vs opus 5. This shows a tier-comparison intent, not a volume or rank. - gpt-6 luna · Google Suggest · US · checked 2026-09-23T22:28:00Z
The seed returned itself first, then gpt 6 luna vs terra and gpt 6 luna price, among others. Pricing and tier comparison are the returned intents. Intent signal, not a volume measurement.
Who this helps
- developers choosing a GPT-6 tier for an agent or coding pipeline
- teams moving off GPT-5.6 Sol or Luna and costing the change
- ChatGPT and Codex users wondering which new model they can access
- founders comparing frontier API prices across vendors
What Sol and Luna are
OpenAI introduced GPT-6 Astra earlier in September 2026 as its top model. The GPT-6 Sol and Luna announcement, which reached Hacker News on September 22, 2026, adds two cheaper tiers. OpenAI says it trained them with methods similar to Astra's so the gains in professional work, factuality, coding, and computer use carry over to faster, lower-cost models.
The naming carries over from GPT-5.6: Sol is the mid tier and Luna the low-cost tier. Search completions also show people looking for a GPT-6 Terra. OpenAI's announcement names only Astra, Sol, and Luna, and we found no GPT-6 Terra in it.
The three tiers side by side
These figures come from OpenAI's API model pages as checked on September 23, 2026. Prices are per million tokens.
- GPT-6 Astra (gpt-6-astra): 10 input, 1 cached input, 50 output. 1,050,000 context, 922,000 max input, 128,000 max output, knowledge cutoff April 30, 2026.
- GPT-6 Sol (gpt-6-sol): 2 input, 0.20 cached input, 10 output. 1,050,000 context, 128,000 max output, knowledge cutoff April 20, 2026.
- GPT-6 Luna (gpt-6-luna): 0.10 input, 0.01 cached input, 0.50 output. 1,050,000 context, 128,000 max output, knowledge cutoff May 18, 2026.
- On all three, prompts over 272,000 input tokens are billed at 2x input and 1.5x output for the whole request, Batch and Flex are half price, and fast mode is double.
What the vendor benchmarks say, and what they leave out
OpenAI reports that on AutomationBench, a business-workflow test run by Zapier, GPT-6 Sol at xhigh effort scored 33.2 percent at 0.27 US dollars per task, above Claude Opus 5 at max effort (26.9 percent) at about 11 times the cost per task. On DeepSWE 1.1 it reports 68.8 percent for Sol at max effort and 66.6 percent for Luna at max effort. On OSWorld 2.0 offline it reports 60.5 percent for Sol at xhigh effort against 60.3 percent for Claude Opus 5 at medium.
Two cautions. First, these are vendor-reported, and cost-per-task figures depend on effort settings you may not use. Second, every Claude comparison in the announcement is against Opus 5 or Fable, not Claude Opus 5.5, which Anthropic released the same day at 4 and 20 per million tokens. Anthropic's own Opus 5.5 table compares against GPT-6 Astra and GPT-5.6 Sol, not GPT-6 Sol. Neither vendor has published a head-to-head of the two newest models, so any claim that one beats the other is not yet supported by primary sources.
Where you can use them
OpenAI says Sol and Luna are available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users, rolling out gradually. Free and Go users can use Luna in the desktop app. The announcement says they are not yet available in Chat. In the API they are gpt-6-sol and gpt-6-luna. The API model pages list both as not supported on the free usage tier, so API access starts at Tier 1.
How to choose a tier
Start from the cheapest model that passes your own tests, not from the top of the lineup. A reasonable first cut:
- Luna for high-volume, well-defined work: classification, extraction, routing, summarizing short inputs, and first-pass triage. At 0.50 per million output tokens it is twenty times cheaper than Sol on output.
- Sol as the default for coding agents, multi-step tool use, and professional writing. OpenAI positions it as the tier most people should iterate on.
- Astra for the hardest end-to-end work, where a measured quality gap is worth five times Sol's price. OpenAI says Astra still leads on computer use.
- Cache aggressively on any tier. OpenAI says cached input reads are discounted 90 percent and that caching improvements in GPT-6 raise hit rates by default.
If you are moving from GPT-5.6
Moving from GPT-5.6 Sol to GPT-6 Sol halves the list price, from 4 and 20 to 2 and 10. Moving from GPT-5.6 Luna to GPT-6 Luna cuts it from 0.20 and 1.20 to 0.10 and 0.50. OpenAI also says GPT-6 Sol makes about half as many factual mistakes as its predecessor on an internal evaluation built from conversations where users flagged errors. Re-run your own evaluation set before switching; knowledge cutoffs and behavior differ between generations.
Limits of this answer
RepoRadar has not run these models hands-on. Prices and limits come from OpenAI's model pages and announcement as checked on September 23, 2026, and may change. Benchmark numbers are vendor-reported. The announcement page we captured did not show a dateline, so the September 22 date comes from when the launch post reached Hacker News.
A useful next action
Take twenty representative tasks from your workload, run them on gpt-6-luna and gpt-6-sol at the same effort, and record pass rate and cost per task. Move a task class up to Sol or Astra only when the cheaper tier fails it. If you are also considering Claude, run the same set on claude-opus-5-5, because no published head-to-head exists yet.
Sources checked
- OpenAI: Introducing GPT-6 Sol and Luna ↗ checked · vendor announcement, global
Primary source. Sol and Luna trained with methods similar to GPT-6 Astra; API prices of 2 and 10 for Sol and 0.10 and 0.50 for Luna, 50 percent below GPT-5.6 promotional pricing; AutomationBench, DeepSWE 1.1, and OSWorld 2.0 offline results with competitor comparisons against Claude Opus 5 and Fable; factuality claim of about half as many mistakes for Sol; 90 percent discount on cached input reads; availability in ChatGPT Work and Codex, Luna for Free and Go in the desktop app, not yet in Chat; API IDs gpt-6-sol and gpt-6-luna.
- OpenAI API docs: GPT-6 Sol model page ↗ checked · vendor documentation, global
Primary source for Sol pricing (2 input, 0.20 cached input, 2.50 cache writes, 10 output), the 272K-token long-prompt surcharge, Batch, Flex, and fast mode pricing, the 1,050,000 context window, 128,000 max output, April 20, 2026 knowledge cutoff, and free tier not supported.
- OpenAI API docs: GPT-6 Luna model page ↗ checked · vendor documentation, global
Primary source for Luna pricing (0.10 input, 0.01 cached input, 0.125 cache writes, 0.50 output), the 1,050,000 context window, 128,000 max output, May 18, 2026 knowledge cutoff, and free tier not supported.
- OpenAI API docs: GPT-6 Astra model page ↗ checked · vendor documentation, global
Primary source for Astra pricing (10 input, 1 cached input, 50 output), 1,050,000 context window, 922,000 max input tokens, 128,000 max output, and April 30, 2026 knowledge cutoff.
- Anthropic: Introducing Claude Opus 5.5 ↗ checked · vendor announcement, global
Used only to show that Claude Opus 5.5 launched on September 22, 2026 at 4 and 20 per million tokens, and that Anthropic's comparison table lists GPT-6 Astra and GPT-5.6 Sol rather than GPT-6 Sol.
RepoRadar separates factual source claims from analysis. Recheck vendor docs before purchase, deployment, or policy decisions.