Composite table for the four benchmarks where Qwen has shown both 3.6-Max (Preview) and 3.7-Max. The progress is not exactly dramatic, but it is significant for 1 month. …Except NL2Repo. Is this real? They claim to have matched Opus in the one thing Opus is hyped for.
@teortaxesTex @Elaina43114880 Am I dreaming? 60+ in SWE bench pro?
Many users are impressed by Qwen 3.7-Max's strong benchmark results on CritPt and similar tests because the gains show practical performance improvements and suggest Alibaba could lead the AI race.
In other notes, Qwen 3.7-Max is the strongest Chinese model on CritPt, and in fact stronger than Gemini 3.5 Flash and Opus 4.6/4.7. The leap from the last generation is almost 4x, the largest I've ever seen.
qwen 3.7 max scores are live on Artificial Analysis !
Conflicted feeling. Binyuan and Junyang are out; I liked them, lamented their departure. But after 3.6 and 3.7, I can see Qwen as a formidable lab for the first time. And yet, Qwen 3.5 is *their* base. Their recipe. This is mostly Alibaba going ham with RL. Almost stolen valor.
Composite table for the four benchmarks where Qwen has shown both 3.6-Max (Preview) and 3.7-Max. The progress is not exactly dramatic, but it is significant for 1 month. …Except NL2Repo. Is this real? They claim to have matched Opus in the one thing Opus is hyped for.
impressively, it does so in fewer tokens than V4-Pro Alibaba has caught up in hard reasoning? no idea about costs
In other notes, Qwen 3.7-Max is the strongest Chinese model on CritPt, and in fact stronger than Gemini 3.5 Flash and Opus 4.6/4.7. The leap from the last generation is almost 4x, the largest I've ever seen.
as an aside, > PNG 17277 × 10523 > 6,2 MB on disk why does Qwen insist on doing this?
Composite table for the four benchmarks where Qwen has shown both 3.6-Max (Preview) and 3.7-Max. The progress is not exactly dramatic, but it is significant for 1 month. …Except NL2Repo. Is this real? They claim to have matched Opus in the one thing Opus is hyped for.
@teortaxesTex Alibaba might win the whole AI race, at this pace of release now every 1 month
@teortaxesTex So they altered GLM's official scores? Interesting. It was definitely not as good as the benchmarks suggested.
@teortaxesTex The new leadership is cooking
@teortaxesTex They've been losing talent left and right to Xiaomi and other labs. This is totally unexpected the numbers are really really good and look promising
@teortaxesTex nice, great to see Qwen getting some frontier wins. i love them based on their consumer GPU open releases -- would love to see them grab some API cash.
@teortaxesTex impressive, critpt isnt very benchmaxxable
@teortaxesTex it's kind of insane how well this model is performing against it's priors. I'm really happy to be seeing these numbers. I need to run it on my internal advertising evals.
@zhenbailian nah it's from qwen 3.6 blog post they revised it down in 3.7, not sure why
@teortaxesTex open models apparently accelerating
ai (ex-crypto) bros: AGI over profits
real-world economics eventually coming home to roost
ai (ex-crypto) bros: surprised pikachu face?
@teortaxesTex Looks like they dumped gated deltanet
@teortaxesTex seems like Qwen's focus on frontier performance is paying off.
@teortaxesTex What's the pricing leap?
@teortaxesTex Well said. Performance gains like this are where DeepSeek starts to feel very practical. Related to that, I am testing a unified AI workflow here: https://aevrynai.com/register?invite=uDaiK5WU
@teortaxesTex It's out on artificial analysis btw
@teortaxesTex Wowzer!
@teortaxesTex What’s critpt
Composite table for the four benchmarks where Qwen has shown both 3.6-Max (Preview) and 3.7-Max. The progress is not exactly dramatic, but it is significant for 1 month. …Except NL2Repo. Is this real? They claim to have matched Opus in the one thing Opus is hyped for.
@teortaxesTex @Elaina43114880 Am I dreaming? 60+ in SWE bench pro?