
OpenAI's GPT-5.6 Luna (Max) is a Game Changer
So according to both:
(Which seems to be fairly trustable) - GPT-5.6 Luna (Max) is ranked:
- #10 in "Intelligence"
- #2 in "Speed"
- #1 (Cheapest) in "Cost Per Task" - 5x cheaper than Deepseek V4 (which is already pretty cheap)
- #12 in "Agentic Index"
- #14 in "GPQA (Diamond)" - Above 90% still
From my own personal experience, even on tasks that take up to 5 minutes, I'm only getting a max usage of about 13 AIC per task. Whereas Claude Sonnet 5 could use around 386 AIC for a similar task and similar timeframe.
The code quality seems pretty good too. I really thought it'd be bad for the cost, but honestly I'm super impressed with this model.
It almost reminds me of the premium request days. I even keep refreshing my billing page to see if it's accurate because I'm surprised with the low AIC usage.
Even here https://benchlm.ai/compare:
- It ranks #23, whereas Sonnet 5 ranks #36 (and Luna is about 10x cheaper)
While I'm regular Claude user - and in most benchmarks Opus 5 (max) is usually #1 in all categories - I'm very impressed with OpenAI for Luna and it's ability to compete with even frontier models, at a fraction of the price.
Curious what others think? Or if anyone else has tried it?
Edit: If anyone knows any other & better benchmark sites please let me know as well!