Skip to content
MustHave.ai
MustHave Hoodie Seal
THE MUSTHAVE NEWSROOM

AI news

What’s new. What matters. What do you think?

Top stories

By member votes

China Telecom Opens Xing4.0-29B-A4B Coding Agent With 4B Active Parameters

Xing4.0-29B-A4B is China Telecom's Apache-2.0 coding model with 29B total parameters, 4B active per token and a native 256K context window.

Microsoft Foundry A2A Tool Reaches GA for Agent-to-Agent Calls

Microsoft Foundry A2A Tool now supports generally available A2A 1.0 calls, authenticated endpoints and cross-agent delegation with defined limits.

Microsoft Gives AI Agents Managed Browsers Through Playwright Workspaces Remote MCP

Playwright Workspaces Remote MCP gives AI agents hosted browsers, 22 tools and evidence capture, but the service remains a preview without an SLA.

Qwen-Image-2.1 Adds Native Transparency, but Its License Blocks Commercial Use

Qwen-Image-2.1 generates transparent 2K images and edits up to 10 references, but its research license bars commercial use without permission.

Databricks Anthropic Deadline Passed for IL5 Workspaces: Verify Endpoints Now

Databricks said Anthropic models would be removed from certain AWS GovCloud DoD IL5 workspaces on September 19. Operators should verify rather than assume.

NVIDIA Is Phasing Out GenAI-Perf: How to Migrate LLM Benchmarks to AIPerf

NVIDIA identifies AIPerf as the successor to GenAI-Perf. This guide explains the migration, traffic models, metrics and client bottlenecks to verify.

Databricks RADAR Targets Silent AI Failures With Streaming Anomaly Detection

Databricks RADAR combines reliability metrics, streaming anomaly detection, alerting and root-cause analysis. Its reported gains still need reproduction.

OpenAI Australia Youth Safety Blueprint Arrives Before a September 22 Policy Deadline

OpenAI published an Australian youth-safety blueprint as a government consultation nears its deadline. Here is what is shipped, proposed and not yet law.

AWS Says Kiro Crew Merged 1,000 PRs in a Week, but Governance Is the Real Test

AWS says Kiro Crew helped merge 1,000 pull requests in seven days. The reusable lesson is its scaling ladder, permission model, audit trail and cost boundary.

Step 5: Preview Weights Are Available: Check Artifacts, Dates, and License

Step 5 Preview now has inspectable weights and serving configurations. This update corrects the earlier release forecast and explains the StepFun license boundary.

Grok Voice Transcribe 2.0 Launches at $0.10 an Hour, but Pin the Model

Grok Voice Transcribe 2.0 adds batch and streaming transcription, diarization, eight channels and low hourly pricing. One version discrepancy matters.