ITmedia AI+ is aggregating community discussion about a model called "Ox Alpha" - an apparently unreleased AI system generating significant conversation in Japan's AI community. Stealth models are a documented phenomenon: some of the most significant capability jumps have been previewed to limited audiences months before public announcement. Whether Ox Alpha is the next one is the question worth watching.
25/08/2026 à 16h27 9 11
An internal OpenAI model reportedly solved ten open problems in mathematics and computer science. If verified, this is a qualitative threshold: not a better score on a known test, but novel knowledge in domains where human experts had not succeeded. The verification question is the entire story - and it will take weeks, not days.
25/08/2026 à 16h27 7 8
Google releases a technical report for DiffusionGemma - applying diffusion model principles to the Gemma family. A research direction that challenges the architectural orthodoxy of autoregressive LLMs gets a production-scale implementation.
20/08/2026 à 22h30 9 11
After weeks of benchmark videos, Kimi K3 is generally available. Now the real test begins.
19/08/2026 à 09h34 12 13
ByteDance has published a technical report on its Douyin multimodal embedding model. The architecture choices reveal how a content platform trains representations differently from a research lab.
19/08/2026 à 09h33 9 10
Anthropic releases a new evaluation framework built to test abstract reasoning rather than pattern-matching to training data - a direct response to the MMLU-saturation problem
18/08/2026 à 09h30 10 12
GLM-5.3 from Chinese lab Zhipu AI targets the coding and cybersecurity verticals specifically, claiming 2× improvement on SWE-Marathon benchmarks and topping CyberGym evals. The vertical specialization race is accelerating.
14/08/2026 à 18h58 6 7
WeLM, the model powering WeChat's AI agent, has quietly scaled to 617 billion parameters with an undisclosed decoding mechanism. No benchmarks, no papers - just a billion-user deployment.
13/08/2026 à 12h57 9 10
The largest open-weight model yet from a Chinese lab. At 2.4T parameters and 1M context, Qwen 3.8 Max enters the tier that was previously proprietary-API-only territory.
10/08/2026 à 16h31 8 10
Meta's open-weight play just got concrete: a 30B model that runs locally, handles multi-step agent tasks, and doesn't route your code through third-party APIs.
10/08/2026 à 16h29 13 13
La scène IA de 2026 n'est plus un seul peloton. Deux dynamiques parallèles s'affrontent avec des critères de victoire différents - et confondre les deux fausse l'analyse.
10/08/2026 à 16h29 6 6
A position paper published at OpenReview challenges a core assumption of the AI scaling thesis: that reasoning capability improves indefinitely with scale. The argument isn't new - but the framing is sharper than usual.
08/08/2026 à 09h58 15 14
