Mercor raises $750M to expand enterprise AI evaluations
Mercor is expanding enterprise AI evaluations after its Series D. Here is what buyers can assess, and what its small internal agent test actually shows.
AI Agents
ERNIE 5.5 Flash Preview opens free API access, with weights still pending
Baidu's ERNIE 5.5 Flash Preview is available through Qianfan and a free OpenRouter endpoint. Check access limits, tool use and the pending weights.
IFM releases PAN 2 world-model weights and training code
MBZUAI’s PAN 2 pairs language planning with video simulation. Here is what is open, what hardware it needs and where its plans still fail.
ElevenLabs launches Scribe v3 for recorded and live transcription
Scribe v3 adds recorded and live transcription upgrades. Here are the speaker-label revisions, turn settings, accuracy caveats and retention checks that matter.
AlphaProof Code verifies Signal code, not every security claim
Google DeepMind used AI agents and Lean to verify specified Signal code properties. Here is what the proof covers, what it assumes and why the limits matter.
Meta launches Muse Ember for cheaper agent tasks
Muse Ember offers low cost agent execution, with threefold token prices above 200K input tokens. We examine access, benchmarks and deployment tradeoffs.
ElevenLabs launches Campaign Manager for outbound AI agents
ElevenAgents Campaign Manager brings outbound phone and WhatsApp workflows, with important limits around callbacks, contact suppression and retention.
LG opens free K-EXAONE 2.5 API with noncommercial limits
LG's free K-EXAONE 2.5 API offers long context and always on reasoning, but commercial use, third-party access and data consent need careful checks.
QwenWork 2.0 adds team agents and event triggers
QwenWork 2.0 adds shared team agents, event triggers and remote task control. Check access, automation limits and China site pricing.
Hermes Security adds AI code reviews, with testing limits
Nous announces Hermes Security for repository reviews. Check test permissions, limits on sharing code and how to assess its evidence before using it.
Feynman update keeps research agents on the chosen provider
Feynman 0.5.66 stops silent provider switching and lets running researchers finish. Here is what to check before upgrading the research agent.
Microsoft Semantic Kernel 1.82 tightens AI tool access
Microsoft's Semantic Kernel .NET 1.82.0 adds explicit controls for database mutations, binds HTTP checks to connections and changes file interfaces. Existing plugin integrations need targeted upgrade tests.
Microsoft Agent Framework 1.25 changes Python and MCP requirements
Microsoft Agent Framework Python 1.25 requires a runtime and configuration review, from MCP discovery limits to workflow sessions and Foundry Toolbox approvals.
Activepieces 0.93.0 updates MCP setup and agent integrations
Activepieces 0.93.0 adds MCP OAuth2 support and setup for more AI clients. Existing workflows still need explicit integration version updates.
Cursor adds voice chat for coding agents, with daily limits
Cursor’s new voice chat lets users discuss coding tasks and hear progress aloud. Account access, daily call limits and privacy settings still matter.
AWS brings Agent Registry to GovCloud, with discovery limits
AWS adds Agent Registry in GovCloud US-West. Approval, registry-wide search access and unverified software references shape how teams should deploy it.
Kanana-2-v-decide brings local AI decisions, with license limits
Kakao's open-weight decision model handles text and images. Its custom license and reference server set important boundaries for deployment.
Google ADK 2.14 adds stronger agent tests, with upgrade risks
Google ADK 2.14 expands agent evaluation. Check breaking changes, scoring across repeated runs and argument matching before upgrading production workflows.
Google adds Memory Bank controls for inspecting agent recall
Google’s November 6 Memory Bank update makes agent memories easier to inspect. Developers still need to plan retrieval scopes, schema changes and retention.
Google Agent Registry skills reach GA with scanning controls
Google makes standalone agent skills generally available, with revision controls and security scanning. Here are the limits and January pricing.
Aina Ring opens preorders for AI agent approvals
Aina puts agent approvals on a ring, with iPhone and Mac requirements, early 2027 shipping and important limits on integrations, privacy and cancellations.
Vercel brings streaming AI cards to Chat SDK
Chat SDK can turn an agent's JSON output into native chat cards. Developers still need to handle platform limits and failures after a partial message appears.
Copilot CLI adds VS Code shortcut and searchable settings
GitHub's stable CLI update brings a simpler terminal-to-editor workflow. Here is how session access, settings and review permissions fit together.
OpenRouter adds Auto Code, with new cost and routing controls
OpenRouter rebuilds Auto and adds Auto Code for coding agents. Check the different cost settings, model restrictions and EU routing limit before switching.