Alibaba sets a date for Qwen's open weights

Alibaba released the model for coding and autonomous work, and says open weights will follow next week. That commitment gives builders a near-term point to compare deployment options.

MiniMax opens H3 with a single-GPU target

Hailuo AI

MiniMax says its open-weight H3 video model can run on a single consumer-class GPU. That hardware target puts local experimentation within reach of a high-end workstation.

Susan Zhang posts an early Qwen3.8-Max read

Susan Zhang

Zhang tested the new Qwen release and compared its formatting style with Sol and its reasoning behavior with Opus. It is an early practitioner report rather than a benchmark claim.

A builder reports lower costs across model suites

Vaibhav Srivastav

Srivastav reports lower costs after changing how his workflow uses different model suites. His post adds an operating-cost data point to the open-versus-frontier discussion.

JFrog traces a critical CVE to a nonexistent SQLite flaw

JFrog Security Research

JFrog documents a critical vulnerability entry for SQLite that describes a flaw the project doesn't have. Once model output enters a formal registry, scanners and downstream teams can treat it as established fact.

AI-assisted COBOL migrations preserved source bugs

arXiv

A study of COBOL-to-Java migration found that generated programs carried bugs from the legacy code into the new implementation. The result tests a common enterprise coding-agent use case against measured migrations rather than a vendor demonstration.

An attribution dispute forms around explorative modeling

Susan Zhang

Zhang alleges that a preprint copied the explorative-modeling pretraining axis and names a specific preprint. The claim remains unresolved, so the linked post records the dispute rather than proving plagiarism.

MCP Apps sends interactive interfaces through the protocol

AI Engineer

The MCP Apps proposal lets servers return sandboxed HTML interfaces whose user events flow back through the host model. Anthropic and OpenAI are contributing to the specification under the MCP steering committee.

MCP Tasks V2 removes the global task list

AI Engineer

Cornelia Davis explains why the durable, stateful V1 design has no production implementations. The proposed V2 makes tasks an optional extension, removes the unfiltered task-list endpoint, and adds explicit update calls.

Anthropic users report unexplained plan usage

Reddit

A user reports unexplained usage and a charge above $480, along with no available support route. The account is unconfirmed, but its date range, billing mechanism, and amount make the claim checkable.

EU AI Act model obligations become enforceable

Euronews

Following yesterday's coverage of the EU AI Act deadline, the obligations for general-purpose AI models are now enforceable. Which model providers will comply with the disclosure requirements remains unresolved.