AM-IA brings together news, research papers and applications relating to artificial intelligence published around the world into a single feed — and makes them available in your European language at the click of a button.
Scroll to browse · Click to filter
The smallest model in DeepSeek's new architecture family relies on an asymmetric division between input and output to compress agent costs.
On September 4, Google opened up its music generation model to the general public and developers: complete songs in 44.1 kHz stereo, richer voices and arrangements, with no free tier.
Gmail Live, Docs Live and Keep Live allow you to query your mailbox, dictate a document or organize notes by voice, on iOS and Android, in English first.
Weights, code and training data are published for the entire range, from 0.9 to 375 billion parameters.
Anthropic moved Claude's computer control to hidden windows on macOS: the agent clicks and enters text while the user continues to work.
The model beats its previous version on all the benchmarks published by Google, and outperforms Claude Opus 5 on three of them.
The same model, two levels of safeguards: prices per token remain stable but hiding costs 75% less, and false safety positives fall sharply.
The expanded partnership between Salesforce and Anthropic installs the data and CRM business rules inside Claude, and makes Claude the default Slack model.
Launched without a publisher's name and hoisted to the top of OpenRouter's charts, the open-weight Ox Alpha model was awarded to Z.ai on August 26. The weights were to follow the next day.
The M5 Ultra combines two M5 Max into a single four-dice package, with 1.2 TB/s of bandwidth. The M6, which is 2 nm, gains 30% GPU compute for AI.
On August 25, Anthropic merged the memories of its two interfaces: what is explained in conversation no longer needs to be repeated when it comes to execution.
Contract review, regulatory monitoring, legal research: the sectoral version of Gemini Enterprise is coming in preview, backed by partner agents.
Designed with Broadcom, the chip has a clear lead over Nvidia Blackwell systems in tokens per user and throughput per kilowatt, according to the InferenceX bench.
At Hot Chips 2026, Nvidia opened the hood of Vera, the Arm processor that will accompany the Rubin GPUs: 88 Olympus cores, eight LPDDR5X controllers and an assumed bias for the IPC.
ChatGPT Work, launched in July at $20 per month, claims 20 million users compared to ChatGPT's billion — and costs OpenAI three times the price of the subscription in tokens.
Called "Thomson" and built on an open basis for $40 million, the model is fed by Westlaw, Practical Law, Checkpoint and Reuters, and deployed in CoCounsel Legal.
Alibaba's video model is moving to general access the day after a $10.2 billion equity placement. It doubles the duration generated and accepts office documents as a source.
Anthropic replaces the message-by-message classification with a reading of the entire thread: the agent chooses between replying, opening a work thread, linking the subject to an existing site or remaining silent.
Computer use, browser use, Skills API and Files API are now generally available. The browser tool does not drive Chrome: it reads the page's accessibility tree.
AWS adds per-query control over sources and time window to its managed search tool, and opens it up to the Ireland and Tokyo regions.