From the archive
AI models are the systems behind chatbots, coding assistants and tools that work with text, images, audio and video. Follow new language model releases, open-weight models, local AI and benchmark results. Our coverage looks at capabilities, context limits, access, pricing and evaluation methods so you can compare the evidence behind a launch, not just the claims in its announcement.
Lead story
AMD has introduced ROCm Hyperloom, an open-source multi-agent harness that profiles, optimises and validates AI inference workloads on AMD Instinct GPUs. AMD says it delivered validated throughput improvements ranging from 1.35x to 7.31x across supported serving frameworks, without human intervention. That makes this more than another software cupboard…
Read the report & discussion ↗Meta AI Watch
Meta has open-sourced Rebalancer, the assignment-problem solver it says has been running across its infrastructure for more than nine years. The useful bit is not simply that another internal tool has escaped into public view: Rebalancer is designed to express and solve the messy allocation problems behind servers,…
Watch Desk
Apple has reportedly expanded its AI hardware range with the M5 Ultra Mac Studio and M6 Mac mini, machines aimed at running open models locally rather than sending every sensitive prompt to the cloud. The practical pitch is lower running costs, faster responses and more control over private…
Watch Desk
Tim Dettmers is launching dlab Open Source Week with tools for running AI locally, building more efficient autonomous research systems and cutting the token bill with a technique called CliffCompaction. His broader argument is that the next wave of AI research should not belong only to frontier labs…
Watch Desk
TypeSafe AI’s Jev edged an open CPU-only alternative by 0.6 percentage points in a new benchmark covering 10,000 decisions across four public datasets. The more useful result is not a knockout victory, but a clearer picture of when a decision model earns its keep and when a small…
Watch Desk
TypeSafe AI has launched Jev, a model designed to return structured decisions and confidence scores rather than generate open-ended prose. The practical promise is a cheaper, faster decision layer for software, although the headline performance claims remain the company’s claims rather than settled independent results.
Microsoft AI Watch
Microsoft’s AI division has published a 37-page code of conduct for increasingly autonomous systems, putting human control, restricted access and honest uncertainty at the centre of its model philosophy. The practical message is refreshingly blunt: more capable AI should not mean less meaningful oversight.
AWS AI Watch
Google and Amazon have topped GlobalData’s latest assessment of cloud-based enterprise AI platforms, as buyers increasingly judge AI services by governance, agent controls and cost management rather than model sparkle alone. The ranking matters because enterprise AI is moving from experimentation into systems that need to be monitored,…
Hugging Face Watch
Hugging Face has published a detailed preview of tokenizers v1, a major rewrite aimed at stopping text preparation from leaving increasingly fast AI systems waiting around. The project reports large gains in decoding throughput, 76% scaling across eight native threads on an Apple M4 Max, and identical token…
AWS AI Watch
Amazon Web Services has made Moonshot AI’s Kimi K3 available through Amazon Bedrock, bringing a Chinese open-weight model to one of the world’s biggest cloud platforms. The important wrinkle is commercial: Moonshot’s licence reportedly requires paid arrangements when companies host K3 as a service above a stated revenue…
DeepSeek Watch
DeepSeek is planning to use more domestic chips to train its AI models, with Huawei expected to begin supplying training processors as early as the fourth quarter, according to The Information. The move would deepen the partnership between two Chinese technology companies trying to reduce their exposure to…
Anthropic Watch
Anthropic chief executive Dario Amodei’s call to “pace the frontier” is being challenged on strategic as well as safety grounds. In a new Stratechery analysis, Ben Thompson argues that slowing model progress could ease the infrastructure, product and competitive pressures now building around frontier AI.
Zhipu AI / Z.ai GLM Watch
Zhipu AI says it will open-source its ZCode programming tool and commission recurring security audits after a controversy over reports that the client uploaded users’ repositories to cloud storage. The important point is that this involves source code, credentials and enterprise data, not merely an awkward privacy setting…
Watch Desk
An independent developer has released two open-weight versions of Boris-2 after uncovering training and data-pipeline mistakes that held the model back. The useful lesson is not simply that another model is now available: the project shows how much performance can depend on unglamorous engineering details, and how sharing…
Unsloth Watch
Unsloth’s 0.1.811-beta release expands Studio from a single-user local AI workspace into a more capable shared tool, adding isolated accounts, AMD support and a preview operating-system sandbox. The practical shift is less glamorous than a new model, but considerably more useful: fewer people having to share one login…
Watch Desk
China is pursuing a different route through the AI race: cheaper, open-weight systems designed to spread widely, backed by coordinated computing infrastructure and overseas partnerships. A Tech Policy Press analysis argues that influence may belong less to whoever builds the most powerful model than to whoever makes its…
Watch Desk
A MarkTechPost comparison of 11 open-source agent harnesses shows that running AI locally is no longer one problem with one obvious answer. The practical choice now depends on whether you value sandboxing, coding power, messaging, hardware flexibility or sheer simplicity.
