01
Overview
OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, 2026, expanding the GPT-6 family beyond the flagship GPT-6 Astra. The new models target a different part of the market: strong professional, coding and computer-use performance at substantially lower API cost. OpenAI says both models were trained using similar methods to Astra and inherit advances in factuality, coding, computer use and alignment. Those capability and benchmark statements are vendor claims unless independently reproduced.
02
GPT-6 Sol and Luna at a glance
GPT-6 Sol is the higher-capability everyday tier. GPT-6 Luna is the faster, much cheaper tier for workloads where unit economics and throughput matter more than maximum capability. GPT-6 Astra remains OpenAI's highest-capability GPT-6 model.
The headline is price. OpenAI lists GPT-6 Sol at \$2 per million input tokens and \$10 per million output tokens. GPT-6 Luna costs \$0.10 per million input tokens and \$0.50 per million output tokens. OpenAI describes both as 50% cheaper than the GPT-5.6 promotional prices it uses for the comparison.
03
What changed from GPT-5.6?
This is not simply a price cut. OpenAI says the GPT-6 training advances behind Astra have been carried into Sol and Luna, with improvements in professional work, coding, computer use and communication style. The company also emphasizes a clearer, less jargon-heavy response style with fewer low-value details.
For developers, the most important practical question is cost per successful task rather than token price alone. OpenAI's launch benchmarks repeatedly frame Sol and Luna around that metric, arguing that stronger results at lower token prices can materially reduce the cost of agentic workflows.
04
GPT-6 Sol pricing
Standard API pricing is \$2 per million input tokens and \$10 per million output tokens according to OpenAI's launch announcement. This makes Sol one-fifth of Astra's launch input and output token prices of \$10/M and \$50/M respectively.
OpenAI's current API docs list GPT-6 Sol at $0.20 per million cached-input tokens and $2.50 per million cache-write tokens, and GPT-6 Luna at $0.01 cached input and $0.125 cache writes. Both models expose a 1.05-million-token context window and 128,000 maximum output tokens. OpenAI applies higher long-context rates once prompt input exceeds 272K tokens, so large-context workloads should use the current pricing table rather than the short-context headline alone.
05
GPT-6 Luna pricing
GPT-6 Luna is unusually inexpensive for a frontier-family model: \$0.10 per million input tokens and \$0.50 per million output tokens in OpenAI's launch pricing table. That makes it especially interesting for high-volume extraction, classification, coding assistance and agent subtasks where a more expensive model would be wasteful.
Low token cost does not automatically mean lowest total application cost. Tool calls, retries, long prompts, latency requirements and failure rates can matter more than nominal token pricing in production.
06
Professional-work benchmarks
OpenAI reports GPT-6 Sol at xhigh effort scoring 33.2% on AutomationBench at about \$0.27 per task. In the same launch table, GPT-6 Astra at low effort scores 30.3% at 3.9 times Sol's cost per task, while Claude Opus 5 at max scores 26.9% at 11.1 times Sol's cost. These are OpenAI-presented benchmark comparisons and should not be treated as independent measurements.
OpenAI also says Luna at high effort improves 5.4 percentage points over its predecessor on AutomationBench while costing 58% less per task.
07
Coding performance
On FrontierCode 1.1 Main, OpenAI says GPT-6 Sol substantially improves over GPT-5.6 Sol and can match Claude Fable 5.1 at xhigh effort at much lower cost. On DeepSWE v1.1, OpenAI reports Sol at max effort scoring 68.8%, within 1.1 percentage points of Claude Fable 5's reported 69.9% xhigh result, at roughly 80% lower cost per task.
OpenAI reports GPT-6 Luna at max scoring 66.6% on DeepSWE v1.1, comparable in its launch comparison with Claude Opus 5 and Fable 5 at medium effort. Again, these comparisons come from OpenAI's launch materials.
08
Computer use
GPT-6 Astra remains OpenAI's top computer-use model, but Sol and Luna are positioned as cheaper alternatives for browser and desktop agents. OpenAI reports GPT-6 Sol at xhigh scoring 60.5% on an offline OSWorld 2.0 evaluation, close to the 60.3% result it reports for Claude Opus 5 at medium effort, at about 80% lower cost per task.
OpenAI also says GPT-6 Luna at max can outperform GPT-5.6 Sol at medium effort on this evaluation at one tenth of the cost.
09
Which GPT-6 model should you use?
Use Astra when maximum capability matters more than cost. Sol is the more natural default for demanding recurring coding, professional and agent workloads where Astra's price is difficult to justify. Luna is the strongest cost-first candidate for high-volume subtasks and workflows that can tolerate a lower capability ceiling.
A practical production architecture may route difficult cases upward rather than use Astra for every request: Luna for cheap routine work, Sol for harder work and Astra for the small fraction of tasks that genuinely need the frontier tier. Whether that architecture wins depends on your workload and should be measured rather than assumed.
10
Important limitations
OpenAI's launch benchmarks are vendor-reported. Benchmark performance does not guarantee application performance. The current API model IDs are gpt-6-sol and gpt-6-luna. OpenAI documents 1.05M context windows and 128K maximum outputs for both. Pricing and regional-processing rules can still change, so production systems should continue reading the live model and pricing pages rather than treating launch-day details as permanent.
11
FAQ
When were GPT-6 Sol and Luna released?
OpenAI announced them on September 22, 2026.
How much does GPT-6 Sol cost?
OpenAI's launch pricing lists \$2 per million input tokens and \$10 per million output tokens.
How much does GPT-6 Luna cost?
OpenAI's launch pricing lists \$0.10 per million input tokens and \$0.50 per million output tokens.
Is GPT-6 Astra still the best GPT-6 model?
OpenAI says Astra remains its best model across the board. Sol and Luna are positioned around cost efficiency.
Is GPT-6 Sol better than GPT-5.6 Sol?
OpenAI reports substantial gains in coding, professional work and computer-use evaluations, but workload-specific testing is still necessary.
Should I use Sol or Luna?
Choose Sol when capability and reliability justify a higher price. Choose Luna when high-volume economics and speed matter more and the task does not require the strongest model.
Sources
Primary and supporting sources
Use the linked documentation to check current pricing, access and product behaviour. The dates above describe this article; they do not guarantee that every vendor claim remains current.