Every editor has that one clip. The background needs to go but something always looks off. Hair gets chopped. Edges look fake. You look up enterprise solutions and they want a full subscription for one use case and even then it is not guaranteed to work. After Effects background removal is not straightforward. CapCut does an okay job until it doesn't. And when it fails on hair or fast motion you are back to square one. MatAnyone 2 is an open source video matting model that does not just detect where the person ends and the background begins. It checks every pixel in the cutout and fixes the ones that are wrong. Hair strands, moving fabric, fast motion & it handles the details that make other tools look amateur. Is it a one stop solution for everything? No. But for an open source tool with this level of capability it is absolutely worth a look.

There's a quiet assumption baked into how most people think about AI models. Bigger means better. More parameters means more capable. If you want the best results, you run the biggest thing you can afford. Qwen3.6-27B makes that assumption uncomfortable. It's a 27B dense model, fully open source under Apache 2.0, and on agentic coding benchmarks it beats Qwen3.5-397B — a model nearly fifteen times its size — across every major test. That's not a rounding error or a cherry-picked metric. It's a consistent pattern across SWE-Bench, Terminal-Bench, and frontend code generation. This doesn't mean bigger models are dead. It means the gap between what you can run locally and what only clusters could handle a year ago just got a lot narrower.

Mistral built Leanstral to do something most AI models don't attempt, write formal mathematical proofs that a compiler can verify as correct. Not "pretty sure this is right" correct. Mechanically, provably, no-exceptions correct. That's a narrow use case, and the audience for it is small. What nobody expected was that a model trained on IMO-level math problems and abstract algebra benchmarks would end up running against open-source codebases and finding bugs that testing and fuzzing had both missed. Five of them previously unreported on GitHub.

Every multimodal model you've used has the same basic system. Text goes in one way, images go through a vision encoder first, audio goes through an audio encoder first, and then everything gets handed off to the language model in a form it can work with. The encoders are load-bearing and you don't just remove them.Google actually removed them.Gemma 4 12B takes raw image patches and raw audio waveforms and projects them directly into the same embedding space as text tokens. There is no vision encoder or audio encoder. One decoder handling everything.

MiniMax handed an internal version of M2.7 a programming scaffold and let it run unsupervised. Over 100 rounds it analyzed its own failures, modified its own code, ran evaluations, and decided what to keep and what to revert. The result was a 30% performance improvement with nobody directing each step. That is not a benchmark result. That is a different way of thinking about how AI models get built. M2.7 is now available on HuggingFace with weights you can download and deploy. NVIDIA is offering free API access if you want to try it without the hardware overhead. The license has a commercial limitation worth knowing about, we will get to that.

UNAM runs the largest university in Mexico. Every year, hundreds of thousands of students take an entrance exam that determines whether they get in. This year, for the first time, the whole thing went remote. They deployed a lockdown browser, AI webcam monitoring, and one human supervisor per 150 applicants. The kind of setup that sounds serious on paper. Then the scores came in. Students hitting 100 or above jumped from 3.5 percent in previous years to 16.3 percent this year. At the very top end, scores of 110 or higher went from 0.9 percent to 5.5 percent. Not a small shift. Not noise. A roughly fivefold increase in top scores, in one year, under one new format. An expert commission investigated. Their conclusion: administer the entire exam again, in person, to around 58,000 people. The rector apologized to students who hadn't cheated. They now have to prepare for and sit another exam anyway.

Last year, a UPS cargo plane went down in Louisville, Kentucky. The crew didn't survive. The NTSB opened an investigation, as it does with every major crash, and added the case files to its public docket system, as it also does. Transcripts, data, findings, all of it accessible to anyone who wanted to look. What nobody thought about was the spectrogram. A spectrogram is a visual representation of sound. It takes audio signals, breaks them down into frequencies, and renders them as an image. The NTSB included one in the Flight 2976 docket because federal law prohibits it from releasing actual cockpit voice recordings. The spectrogram felt like a reasonable middle ground, you could see that audio existed without being able to hear it. Then Scott Manley, a YouTuber with a background in physics, pointed out on X that spectrograms encode enough data to work backwards from. The image wasn't just a picture of sound. It contained the sound. People ran with it. Using AI tools, they took the spectrogram and the publicly available transcript and reconstructed approximations of what the cockpit voice recorder actually captured. The voices of two pilots who died in that crash started circulating online. The NTSB shut its entire public docket system down.