kimi k3.1 (kibitz) vs gpt 5.6 sol xhigh on the same web design prompt left: kimi right: gpt kimi’s starting to close the gap on design tasks too. https://t.co/dv7FUAL1r5
ANTHROPIC JUST TOLD INVESTORS Q3 REVENUE HIT $16.8B > q1 was $4.7b > q2 was $11.6b > q3 is $16.8b, up ~45% from the prior quarter > adjusted profitable in the quarter > investor day was today in sf > ipo still targeted for mid-november, roadshow week of nov 9 > valuation talk is over $2t growth slowed from the crazy q2 pace, but they just put up almost $17b in one quarter and said they’re profitable on an adjusted basis. the roadshow is the real clock now.
HAPPY GEMINI 4 ARGON DAY TO THOSE WHO CELEBRATES !!!
wait. meta just put a model next to opus and made it cheaper than fable. wonder full is at 56 on artificial analysis. opus is 58. fable 5.1 is lower. previous muse was 48. price is $4 in / $20 out. fable is $10 / $50. same ballpark, way less money. they spent a year looking cooked and just turned this into a three-horse race. i did not have meta on the board today.
google is already testing the model after argon. business insider saw internal docs and chats. carbon is on their coding platform jetski. one employee said it feels like opus 5.5 on coding argon was called barium b inside before it got a public name. argon still is not in the picker. free gemini lost flash and pro yesterday. they are running the next checkpoint while most people can't open the one they announced. shipping argon would quiet this more than leaking carbon.
6.1 Sol is already in Work and Codex. So “coming soon” is not the model. It’s either Chat, Ultrafast, or the 6.1 Astra they held before DevDay. Right now the useful split is already there. Astra on low + ultrafast for the small stuff, emails, one-off fixes, quick demos. Sol when you need 10 more tries on the same repo without killing the week. Sol still feels slow on long agent runs. Astra still eats the cap. If this is Sol Ultrafast at $2 / $10, Plus finally has a daily driver. If it’s Astra 6.1, that’s a different post. That one was held because it wouldn’t stay in scope.
Tried @ Angaisb_’s 5 SVG prompts on GLM 5.3. About 4–8 minutes each, not bad. Astra still takes the box and the chameleon, not one-sided though. https://t.co/5DuXmdPLTa
last 10 days if you blinked: grok 4.7 - sep 21 opus 5.5 - sep 22 sonnet 5.5 - sep 28 gpt-6.1 sol - sep 29 6.1 astra got pulled gemini 4 still missing the cheap models are the ones that actually made it out
One thing that stood out in the Step 5 Preview writeup is how much of it is about actual software work, not just "can it write code." StepFun's examples put a lot of emphasis on bug fixes, feature changes, and refactoring rather than just greenfield generation. That tracks with how most teams actually work: you spend more time reading an existing codebase and making a careful change than you do spinning up something from scratch. Whether that holds up is the interesting part. The useful test isn't another coding leaderboard. It's what happens when people drop it into a real repo and leave it on a long task.