It’s over. Google’s new image model Nano Banana just killed Photoshop. Say what you want, it edits anything just by describing it — no tools, no layers, no hassle. 10 insane examples you won’t believe: 1/ Asked Nano-Banana for a straight-on portrait — it nailed it instantly.

State of the art is at around 50% on SWE-bench Multilingual and 10% on SciCode. There's still *a lot* more we can do with current model paradigms, and we don't even need new benchmarks to track progress yet.

We have MLE-bench, PaperBench, and OAI PR's, which all cover skills relevant to automating AI research. GPT-5 was an infinitesimal improvement from o3 on all three. Big jumps would be worth paying attention to.

We evaluated GPT-5 in OpenHands and it's the new number one coding agent model for us! Using exactly the same tools and harness it's 1.4 points better than Claude Sonnet 4 at 60% of the price. Full results here: docs.google.com/spreadsheets…

A bit sad how the GPT-5 launch is going so far, especially after the long wait and high expectations - The automatic switching between models (the router) seems partly broken/unreliable - It's unclear exactly which model you're actually interacting with (standard or mini, reasoning level, old/new) - Router issues probably contribute to many negative first impressions ("GPT-5 is bad"), especially when you can't easily tell which model is handling your requests - e.g. GPT-5 Thinking performs better at writing compared to standard GPT-5 ("think harder" prompts might also help) - I'm seeing lots of complaints about legacy models being removed and usage limits for Plus (and also Free) accounts ("bring back 4o, o3, ...") - Context window lengths aren't clear, nor how many and which exact GPT-5 models ChatGPT uses (standard, mini, new, old) - no indicator of which model is active at any moment - The gradual rollout and current capacity constraints add to confusion, because not everyone can access the new models yet - Mixed-up charts during the livestream provided plenty of free material for chart-crime and vibe-charting posts, which hasn't helped build trust or confidence - And of course, the community had extremely high expectations across all fronts and modalities, as usual I'm hoping improvements in the next hours/days, and today's AMA, can clarify some of these open issues and questions

We have a long history of using games to measure progress in AI. 🎮 That’s why we’re helping unveil the @Kaggle Game Arena: an open-source platform where models go head-to-head in complex games to help us gauge their capabilities. 🧵

New release from WeirdML! We ran all the major historical models we could find, going back to GPT-4 in June 2023! We see that the cost/capability frontier moves forward significantly every 6 months. We also looked at the gap between open and closed models (see next tweet!).

Introducing NEO: The first Agentic Machine Learning Engineer. It works like a full-stack ML engineer that never sleeps: handling data exploration, feature engineering, training, tuning, deployment, and monitoring, end to end. Powered by 11 specialized agents, NEO runs autonomously, saving ML engineers thousands of hours and making them 10x faster. Benchmarks: Tested on 75 Kaggle competitions, NEO scored a medal in 34% of them, significantly outperforming Microsoft’s RD Agent (22.4%) on the MLE Bench. This sets a new state of the art for autonomous ML systems. NEO runs on a novel multi-agent orchestrator, powered by a multi-step reasoning engine, context transfer protocol, and agent memory, built to solve complex workflows end to end. And with human-in-the-loop mode, you can guide, inspect, and override any step. You're always in command. NEO is built for real world workflows and ready for production. NEO is here to make every ML engineer truly superhuman. Watch NEO in action:

Announcing the Artificial Analysis Music Arena Leaderboard: with >5k votes, Suno v4.5 is the leading Music Generation model followed by Riffusion’s FUZZ-1.1 Pro. Google’s Lyria 2 places third in our Instrumental leaderboard, and Udio’s v1.5 Allegro places third in our Vocals leaderboard. The Instrumental Leaderboard is as follows: 🥇 @SunoMusic V4.5 🥈 @riffusionai FUZZ-1.1 Pro 🥉 @GoogleDeepMind Lyria 2 @udiomusic v1.5 Allegro @StabilityAI Stable Audio 2.0 @metaai MusicGen Rankings are based on community votes across a diverse range of genres and prompts. Want to see your prompt featured? You can submit prompts in the arena today. 👇 See below for the Vocals Leaderboard and link to participate!