The short answer
Claude Sonnet 5 vs GPT-6 Luna: specs side by side
| Claude Sonnet 5 | GPT-6 Luna | |
|---|---|---|
| Made by | Anthropic | OpenAI |
| Released | June 30, 2026 | September 22, 2026 |
| API model ID | claude-sonnet-5 | gpt-6-luna |
| Context window | 1M tokens | 1.05M tokens (922K input) |
| Max output | Not published | 128,000 tokens |
| Knowledge cutoff | Not published | May 18, 2026 |
| Inputs | Text, images and files | Text and images |
| API price | $2 input / $10 output per 1M tokens | $0.10 input / $0.50 output per 1M tokens |
| In AimiChat | Balanced quality · Aimi+ ($15/month) | Fast quality · Free ($0) |
Which to use for each task
Quick answers and simple rewrites
Luna is built for speed and is free.
Writing you will send or publish
Sonnet 5 needs fewer edits and keeps to style rules.
Coding help
Sonnet 5 explains and fixes code more reliably, especially beyond a few lines.
Recent facts without web search
Luna's knowledge runs to May 2026.
What to know before you choose
A practical pattern: draft or explore with GPT-6 Luna, then switch to Claude Sonnet 5 in the same chat when the answer needs to be right or polished.
In AimiChat, GPT-6 Luna is on the Free plan and Claude Sonnet 5 is available from the Aimi+ plan.
Published benchmarks
Each company reports different tests with different settings, so these are listed per model rather than head to head.
| Model | Benchmark | Score | Reported by |
|---|---|---|---|
| GPT-6 Luna | DeepSWE v1.1 (max effort) | 66.6% | OpenAI launch post, as reported by VentureBeat and Codersera |
Try both in AimiChat
Claude Sonnet 5 runs on the Balanced quality and GPT-6 Luna on Fast. On a plan that includes both, you can send the same prompt twice in one chat, switching quality in between, and compare the answers directly. Read more about Claude Sonnet 5 and GPT-6 Luna, or see which plan includes which model.
Sources
Specifications and prices come from the model makers and API listings below. Benchmark figures are reported by the company that made the model and are not independent tests.