Open Source - Page 30 of 33
Open-source AI projects, community innovations, collaborative development, and freely accessible AI tools and frameworks.
642 articles
View complete article list
Apple’s STARFlow-V Generates Video Without Diffusion, but Long Sequences Falter
Apple’s STARFlow-V sidesteps diffusion entirely, a bold move in an era dominated by noisy, iterative generation. Yet for all its ingenuity, the model still stumbles on the long haul.
WIRED Roundup: DOGE Survives, Facebook Dating Launches, Amazon Chases AI Lead
This week, the uncanny valley just got deeper. Zoe Schiffer and Leah Feiger break down three stories that refuse to follow the narrative.
ServiceNow AI releases Apriel-1.5-15B-Thinker, an open-weight coding model
The model is called Apriel-1.5-15B-Thinker. Fifteen billion parameters, open-weight, and built by ServiceNow AI to do something refreshingly rare in the coding-AI race: it thinks before it types.
Mistral Large 3 Shows Superior Collective Performance in Benchmark Tests
Another open-source model claims it can compete with the giants. This time, it might be right. Mistral Large 3 is now the second-ranked open-source model on the LMArena leaderboard, sitting sixth overall.
Google AI model mimics smartphone sharpening in ferry boat image
Look closer at that ferry boat image. The overly sharp edges, the artificial pop, it’s the unmistakable fingerprint of a smartphone camera. Google’s AI has learned to mimic that look, and it’s doing so with unsettling precision.
AI tool Suno lets producers generate full country tracks from a genre prompt
Country music's production line just got an assembly robot. The new AI tool Suno can generate complete, genre-faithful tracks from a simple text prompt, sidestepping the traditional grind of studio sessions and session musicians.
Mixture-of-Experts AI Models Run 10× Faster on NVIDIA Blackwell NVL72
Mixture-of-Experts models are compute gluttons. Running them is famously expensive. NVIDIA’s new Blackwell NVL72 system directly tackles that cost, executing models like DeepSeek-R1 ten times faster than its Hopper predecessor could.
Fortell hires NYU Langone researchers to scientifically test its hearing aid
Fortell just threw its new hearing aid into a scientific gladiator arena. The startup enlisted audiology and neuroscience experts from NYU Langone Health to design a blind test, pitting its device directly against the reigning champion: Phonak.
Arcee releases Apache-2.0 Trinity Mini (26B) and Nano Preview (6B) models
Arcee just dropped a gauntlet. Two new models, both under Apache 2.0, a license that signals real openness. Trinity Mini, with 26 billion parameters but only 3 billion active per token, is built for speed and reasoning.
Mistral launches Large 3, an Apache-2.0 open-source model for language, images
Mistral's latest model has arrived, and it's free for anyone to use. The company just launched Mistral Large 3 under the Apache-2.0 license. It processes both text and images.
DeepSeek Math V2 Deploys Dual-Component Architecture, Two-Stage Verifier Training
Ask an AI for a math answer today, and you'll get one instantly. Ask it to produce a formal proof—the structured, logically sound kind required for publication—and the task becomes fiendishly difficult.
SAM3D isolates specific items—like a tall lamp—beyond broad-class segmentation
Imagine pointing at a 3D scan and asking for just the tall lamp beside the sofa, not the sofa, not the floor lamp, not everything vaguely vertical. Existing models can’t do that. They chunk the world into broad bins: chair, table, human.
DeepSeek launches V3.2 reasoning model, Speciale API available until Dec 15 2025
DeepSeek’s V3.2 reasoning model arrives with a bold claim: it’s designed to be your daily driver at GPT-5 performance.
Pinokio 5.0 adds Dev Mode with one-click terminal launch for AI scripting
Most software tries to hide the terminal. Pinokio 5.0 just makes it useful. The new version shoves command-line tools directly into the app itself. You can launch them with a click, bypassing the usual mess of separate windows.
DeepseekMath-V2 Generates and Verifies Proofs, Aiming to Pop US AI Bubble
American AI labs treat their best models like state secrets. Deepseek just published the blueprint for theirs. The Chinese company's latest math model, DeepseekMath-V2, solves gold-medal International Math Olympiad problems.
Qwen3-VL scans two-hour videos, hits 96.5% on DocVQA, 875 OCRBench
Every AI model launch is a marketing campaign, and the benchmark scores are its slogans. Alibaba's new Qwen3-VL is selling a simple, brutal pitch: it can watch a two-hour video and tell you what happened.
DeepSeekMath-V2 Wins Gold at IMO 2025, Tops China Math Olympiad
DeepSeekMath-V2 just aced the hardest math tests on Earth. Gold at the International Mathematical Olympiad 2025, a top score on the China Mathematical Olympiad, a 118 out of 120 on the Putnam. It beat the best human Putnam score by a wide margin.
SAM3 uses concept segmentation to locate any object described in images or video
Image recognition software has always needed a menu. Ask it to find a dog, a car, a person. Ask it to find "my grandmother's chipped teacup" or "the kid in the blue hat" and it fails. The system is blind to anything it wasn't explicitly told to see.
Startup Aims to Reinvent AI Interconnect, Tackling the Traffic Jam
The data highway is clogged. As AI models balloon to trillion-parameter scale, the pipes connecting chips, clusters, and continents have become the bottleneck, a crawling traffic jam that throttles training speed and inflates costs.
Meta Eyes Google TPUs, Up to One Million Units, as NVIDIA Alternative
Mark Zuckerberg is about to spend a breathtaking sum of money. Meta's capital expenditure is projected to blast past $100 billion in 2026.
Featured Resources & Reviews
No Code MBA Course Review
Learn to build AI-powered apps without coding. Our comprehensive review of No Code MBA's course.
AI Tools & Resources
Curated collection of AI tools, courses, and frameworks to accelerate your AI journey.
Weekly AI Digest
Get the week's most important AI news delivered to your inbox every week.