What Happened
OpenAI introduced BioBasicsBench, a 100-task benchmark that tests how AI agents interpret raw experimental data and handle everyday wet-lab science tasks [1].
Why It Matters to Businesses
For teams building scientific assistants, this is a more relevant evaluation target than general-purpose question answering: the work starts with experimental evidence, not a polished prompt.…
What Happened
LiteLLM 1.108.x LTS expanded provider and model support, routing controls, MCP gateway capabilities, budget reporting, and telemetry. It also includes fixes for tool calls, streaming, credentials, cost calculations, retries, and database behavior. The subsequent 1.108.1 and 1.108.2 patches backported fixes; 1.108.2 also changes how unsupported hosted MCP tools are handled through translated routes:…
What Happened
Recent listings describe a macOS terminal for AI coding agents, a local AI-powered launcher and file finder, and on-device dictation for Mac apps and agents [1][2][3]. Others describe a desktop AI companion, an SEO/GEO system for people and agents, and AI assistance for production issues [4][5][6].
These are product descriptions, not confirmed launch…
What Happened
AWS announced updates across AI, data, infrastructure and media services on November 6–7, 2026. The changes most relevant to production AI systems are:
Models and agent governance: Moonshot AI’s Kimi K3.1 Pro became available on Amazon Bedrock on November 7. Its Highspeed option supports coding and document-processing workloads; Bedrock also offers explicit prompt…
What Happened
New model releases put cost, deployment choice and evaluation back in focus. Thinking Machines launched Inkling 2 Small, a multimodal mixture-of-experts model, but said its weights would follow external safety testing. Moonshot released open-weight Kimi K3.1 Pro, while other open-model releases expanded the options for teams considering self-hosting [2][3].
Agent results were less…
What Happened
Two recent analyses highlight attacks that can bypass familiar security assumptions. A review of Binance’s October 2026 incident reports that attackers used social engineering, an internal AI tool, stolen credentials and access to a wallet-signing service to obtain valid signatures without stealing private keys. An EIP-7702 delegation then gave the attacker persistent wallet…
What Happened
GitHub is extending Copilot beyond code completion. @GitHub is generally available in Slack and Teams for investigating bugs, creating issues, implementing changes, and opening pull requests from conversations. Teams adds a shared Canvas for reviewing plans and changes; new issues and pull requests retain attribution to the originating conversation and user. [4]
Copilot’s…
What Happened
Recent papers point to a practical pattern: improving a model is only part of improving an AI application. The surrounding evaluation, permissions, data and inference systems can determine whether gains hold up in use.
Agent capability can introduce new failure modes. In SEABench, agents allowed to modify their own instructions, memory protocols or…
What Happened
Recent releases put more emphasis on execution control and reliability than on new agent patterns. Claude Code now asks or denies when its server safety check is unavailable in auto mode, rather than silently falling back to a built-in classifier. It also closed sandbox and permission bypasses and made MCP tool calls background…
What Happened
Posit Assistant now exposes estimated cost per turn and conversation, including subagent work. Turn details show requests, tokens, time and cost; usage views distinguish a conversation from the running session, which may span multiple conversations. Token views also show cache reads, writes and cache efficiency. These are estimates, not a substitute for the…
What Happened
Recent AWS releases show two distinct infrastructure priorities: using expensive GPUs more efficiently and making agent deployments easier to govern. SageMaker AI now supports NVIDIA Multi-Instance GPU (MIG) partitions on certain real-time inference-component endpoints, allowing multiple appropriately sized models to share one device with hardware-isolated capacity. Partitioning does not reduce the instance’s hourly…
What Happened
The developments documented here are primarily llama.cpp inference-engine changes, not new open-weight model releases. They expand hardware support and address operational faults across its CPU, GPU, and NPU backends. Hexagon gained a 3D convolution path that reuses its tiled 2D implementation; OpenCL gained additional data types for the repeat operation; and WebGPU gained…