I think this is a completely different class from Luna and is much more akin to 6.1 Sol in terms of capabilities and general intelligence.
replyMaybe, although google tends to benchmaxx. If it's actually sol for free that's kinda crazy good.
replyIt's definitely not the case that small models only do well in benchmarks. We're working on a battle system on top of OpenCode, so have data for real people doign real work.
replyRight now DeepSeek Flash 4.1 Flash has a 64% win rate against GPT 6.1 Sol. Small data still, but it has 33 wins vs. 18 losses, which is confidence interval of 52% - 78% win rate. https://agentduel.com/leaderboard/models/deepseek/deepseek-v...
That's coding though. I was talking about general usage.
replyOkay, makes sense, though I don’t see a huge difference in quality between Luna and Astra for basic chat usage.
reply3.8 flash was easily the best general everyday person model yet though
reply3.8 flash was never served to free users on google though?
reply3.8 Flash was served to free users in AI Mode IIRC
replyOnly for pro and ultra users
replyThe free default was 3.5 flash lite.
Ah, you're correct.
replyMaybe I’m spoiled but I found it repulsively dumb. Some of the highest rates of hallucinations since GPT-3. Incredibly lazy. Heavily biased to parametric knowledge and speed over accuracy. I would rather have no AI at all than use that.
reply