TL;DR: In the GPT-6 Sol vs Luna matchup, Sol is OpenAI’s mid-tier workhorse for complex coding, research and computer use, while Luna is the fast, cheap model for simple, high-volume jobs like classifying requests or pulling fields from invoices. In the API, Sol costs $2 per million input tokens and $10 per million output tokens; Luna costs $0.10 and $0.50, so it’s 20 times cheaper. In ChatGPT, paid plans mostly get Sol, while Luna shows up on the cheaper tiers. GPT-6 Astra sits above both for the hardest work.
OpenAI’s GPT-6 family now has three main tiers, and the names don’t tell you much on their own. If you’re choosing a model for an app, an agent or your own ChatGPT work, the real question is how much intelligence a task needs and how much you want to pay for it. This guide compares GPT-6 Sol vs Luna using OpenAI’s own documentation, with OpenAI’s benchmark claims clearly labeled as the company’s results.
GPT-6 Sol vs Luna at a glance
| GPT-6 Sol | GPT-6 Luna | |
|---|---|---|
| Position | Balanced speed, cost and intelligence | Fastest and most cost-effective |
| Best for | Complex coding, research, computer use, multi-step agents | Focused, repeated tasks with a clear goal |
| API input price | $2 per 1M tokens | $0.10 per 1M tokens |
| API output price | $10 per 1M tokens | $0.50 per 1M tokens |
| API model ID | gpt-6-sol (newer: GPT-6.1 Sol) | gpt-6-luna |
| DeepSWE v1.1 (max effort, OpenAI-reported) | 68.8% | 66.6% |
Prices come from OpenAI’s GPT-6 Sol and Luna announcement and its model guide for the GPT-6 family.
Where Sol and Luna fit in the GPT-6 family
OpenAI introduced GPT-6 Astra first and calls it “the most intelligent and aligned model in the world.” Sol and Luna launched on September 22, TechCrunch reported. OpenAI describes them as cheaper models trained “with similar methods as GPT-6 Astra.” A newer version, GPT-6.1 Sol, arrived at the end of September.
OpenAI’s October 2 model guide lays out the family like this:
- GPT-6 Astra: the hardest reasoning work, at $10 input and $50 output per million tokens.
- GPT-6.1 Sol: “near-Astra intelligence for a fifth of the price,” for complex coding, research and computer use.
- GPT-6 Luna: “fast and efficient everyday work at scale,” such as extracting invoice fields, classifying requests or producing structured summaries.
Price: Luna is 20 times cheaper
OpenAI cut API prices for both models by 50% compared with the GPT-5.6 Sol and Luna promotional pricing. That leaves Sol at $2 and $10, and Luna at $0.10 and $0.50, per million input and output tokens.
Cached input is cheaper still. OpenAI’s model guide lists cached input at $0.10 per million tokens for GPT-6.1 Sol and $0.01 for Luna. If your app reuses the same long instructions or documents on every call, caching can matter as much as the headline price.
Performance: what OpenAI says each model can do
All the scores below are OpenAI’s own results, run in its research environment or through its API. OpenAI notes that production ChatGPT can behave slightly differently.
Coding
On DeepSWE v1.1, a test of long software-engineering tasks in real codebases, OpenAI says GPT-6 Sol scored 68.8% at max effort and GPT-6 Luna scored 66.6%. That’s a small gap for a 20-times price difference, which is why Luna is worth testing for coding agents before you default to Sol.
Business workflows and agents
On AutomationBench, which tests workflows across 47 tools, OpenAI reports that Sol at extra-high effort scored 33.2% at $0.27 per task. OpenAI says Luna at high effort improved on its predecessor by 5.4 percentage points at 58% lower cost per task. On Agents’ Last Exam, Sol at max effort scored 56.4%.
Computer use
On OSWorld 2.0 offline, OpenAI says Sol at extra-high effort scored 60.5%. It also says Luna at max effort beat GPT-5.6 Sol at medium effort for one-tenth of the cost. Astra remains OpenAI’s top model here.
Factuality
OpenAI says Sol makes about half as many factual mistakes as its predecessor on an internal test built from conversations where users had flagged errors. It says Luna at higher effort matches GPT-5.6 Sol “at about a hundredth its cost.” These are internal evaluations, not independent tests.
Which is in ChatGPT, and on which plan?
This part is still shifting as OpenAI rolls GPT-6 out, and its own pages don’t all describe it the same way. Here’s what they say as of mid-October:
- OpenAI’s help center says that when GPT-6 is selected in Chat, GPT-6 Sol powers the Instant, Medium, High and Extra High thinking levels for eligible paid users. Plus includes Medium and High but not Extra High.
- The same help article says Free and Go users get GPT-5.6 Luna in Chat, not Sol.
- The ChatGPT pricing page lists GPT-6 Sol for Plus and GPT-6 Luna for Go and Plus.
- At launch, OpenAI said both models were available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu, and that Free and Go users could use GPT-6 Luna in the desktop app.
In practice, the model menu in your own account is the final word. The help center notes that GPT-6 appears as “GPT-6” in the menu and isn’t labeled Sol. For a fuller view of the plans, see our breakdown of whether ChatGPT Plus is worth it.
When to choose Sol
- Multi-step agents that call tools, browse or use a computer.
- Large coding tasks that need clean, mergeable changes.
- Research and analysis where a wrong answer is costly.
- Work where you’d otherwise reach for Astra but want to spend less.
When to choose Luna
- High-volume jobs: tagging support tickets, classifying emails, extracting fields.
- Structured summaries and routine rewrites.
- Chat features where speed matters more than depth.
- Coding agents on a budget, after you’ve tested quality on your own tasks.
A simple way to decide
- Start with Luna on a sample of real tasks and measure the success rate.
- If it fails on a meaningful share, try Sol at medium effort.
- Raise Sol’s reasoning effort only if quality still falls short. OpenAI’s guide suggests keeping extra-high or max effort only “if the improvement justifies the added time and cost.”
- Reserve Astra for the hardest problems.
That approach follows OpenAI’s own advice to measure task success, latency and “cost per successful task” before deploying.
Related reading
See how OpenAI’s models stack up against Google’s in our ChatGPT vs Gemini comparison, and read about GPT-6’s Intelligent UI in ChatGPT.
The bottom line
Pick GPT-6 Sol when the task is complex, open-ended or expensive to get wrong. Pick GPT-6 Luna when the task is clear, repetitive and high-volume, since it costs a twentieth as much and, by OpenAI’s numbers, isn’t far behind on coding. Test both on your own work, because OpenAI’s benchmarks are a starting point, not a guarantee.
The Teck Hustlers Editorial Team covers technology, business, finance and gaming news for readers who want the story behind the headline. Every article is researched, written and edited by our in-house team, checked against company announcements, filings and reputable reporting, and updated when new facts emerge. We focus on what each development means for founders, professionals, investors and everyday users. Have a tip or a correction? Reach us through our contact page.
