What a day! I couldn't really use the last Luna for much (wasn't smart enough) or Astra (too expensive). So this release is really exciting. I can probably use Sol 6 as much as I want in the week, which as great.

AI needs to get rid of model versioning and model effort combinations. It's like selling an automatic transmission but still asking you to choose the gear, then after the trip telling you how much fuel you burned.

Marketing team must be stoked with 'Sol and Luna.' Let's see if the actual benchmarks live up to the celestial branding.

I've actually been finding gpt5.5-medium much superior for embedded lisp and C programming recently. Maybe less safety nerfed. Feels much clearer, straighter forward.

Because the top-tier model is slow and expensive.

The models below are competing on value for money.

In service development coding, a top-level model is not required.

I hope this kind of competition continues.

hmm, it somehow continued the trend of being basically the same score on https://artificialanalysis.ai/ as the 5.6 variants.

I kind of hated Astra for it's poor instruction following and stopping all the time plus bad code quality. It somehow feels a bit like some of the popular open models but with a lot more knowledge or peek capability. But it doesn't reach peek that often

"GPT‑6 Luna at max effort scores 66.6%, comparable to Claude Opus 5 and Fable 5 at medium effort...." Hmm... coincidental percentage, I'm sure.

I've been seeing numerous reports which compares Astra 3D models on launch day to what they produce now. They seem to have nerfed their model.

Has anyone else noticed this?