| GPT-5.6 Luna | GPT-5 mini | |
|---|---|---|
| Price in/out ($/M) | $0.1 / $0.6 | $0.25 / $2 |
On the spec sheet Luna wins on everything you can count: $0.1 / $0.6 per M tokens against $0.25 / $2, so two and a half times cheaper on input and more than three times on output, with a 1.05M context window against 400K. Both take text and vision. What GPT-5 mini has and Luna does not is a published track record: 82.4% MMLU, 50% on SWE-bench Verified, 1288 Arena Elo, 78% AIME. OpenAI has released no benchmarks at all for Luna. So the honest framing is not which is better, it is how much unmeasured risk the cheapest token in the OpenAI lineup is worth to you. For classification, extraction and other high volume jobs where you can grade the output yourself, Luna is the better economics by a wide margin. For anything where a silent quality drop is expensive, stay on the model with numbers until Luna has some.