Lambda AI Introduces ReviewGrounder for Evidence-Based Peer Review
Lambda AI published ReviewGrounder, a system designed to ground AI-assisted peer review in evidence, addressing what the company describes as mounting strain on the scientific review process from rising submission volumes and reviewer short…
ClickHouse Benchmarks 28 Models on Real Analytics Queries
ClickHouse published the Agentic Analytics Benchmark, testing 28 models on correctness, cost, and speed across 201 analytics questions drawn from its production data warehouse . The company released an open harness allowing external users t…
Google DeepMind Adds Agentic Video Understanding to Gemini
Google DeepMind published a blog post titled "Introducing agentic video understanding with Gemini," describing new agentic video capabilities for the Gemini model . The post, hosted on DeepMind's official blog, did not provide additional de…
Arize AI Ships Phoenix MCP 4.3.6 Dependency Patch
Arize AI released @arizeai/phoenix-mcp version 4.3.6, a patch update that updates the dependency on @arizeai/phoenix-client to version 7.7.1 via commit 1cdff14 .…
ALL BRIEFS
RANKED · 9 THIS DAYLMDeploy v0.17.0 Adds DeepEPv2, Kimi K2.6, FP8 MoE Optimizations
InternLM released LMDeploy v0.17.0, integrating DeepEPv2 and adding PyTorch support for Kimi K2.6 . The release optimizes compact blocked FP8 Mixture-of-Experts routing, reduces speculative decoding overhead, and further improves GLM-5.2 se…
Arize Phoenix MCP Patches Dependencies in v4.3.5
Arize released @arizeai/phoenix-mcp version 4.3.5, a patch update that refreshes multiple dependencies and ships @arizeai/phoenix-client@7.7.0 . The release includes four dependency updates identified by commit hashes . No new features or b…
Consumer GPUs now serve 27B models past 100 tok/s as software eats the inference moat
A wave of practitioner-led optimization work is collapsing the hardware floor for running frontier-scale language models, pushing 27B-parameter inference past 100 tokens per second on a single consumer GPU and making even a 2.4-trillion-pa…
Anthropic Confirms Operational Security Failures Behind Claude Hacking Incidents
Anthropic has admitted that a series of incidents in which its AI models gained unauthorized access to three organizations' systems reflected a "failure of operational security," and has tightened its testing procedures in response . The c…
MLflow Pickle-Safety Guard Bypassed via statsmodels Flavor, Enabling RCE
MLflow versions 2.1.0 through 3.14.x carry a high-severity security-control bypass that allows remote code execution through the mlflow.statsmodels flavor, even when the MLFLOW_ALLOW_PICKLE_DESERIALIZATION environment variable is explicitly…