AI Launch Radar

Opus 4.8 delivers targeted coding gains—4% higher recall, sharply reduced noise, and ~4× fewer unremarked flaws—making it more reliable for code...

Four tools highlight the new infrastructure layer for AI engineering teams:

• Integuru generates reliable backend APIs in minutes, ditching slow...

Microsoft will debut its own coding model at Build next week, integrating it into GitHub Copilot to reduce reliance on OpenAI and Anthropic. The shift...

Meta's AI team announced ATLAS, one of the largest automated formalization efforts to date, targeting improved AI safety and reliability via formal verification. This research push signals long-term progress toward more trustworthy systems.

Three new tools show AI handling coaching, ads, and outbound:

• Firecoach clones playbooks for AI roleplays and scored feedback
• Ava Studio...

Three key AI updates from this week's roundup:

• Microsoft will release homegrown coding models next week to defend GitHub Copilot against rivals like...

AgentDoG 1.5 introduces a lightweight alignment framework that tackles dynamic safety risks in open-world LLM agents through an updated taxonomy, data...

Memory bandwidth, not compute, is becoming AI's key bottleneck for inference. Xcena's MX1 chip moves processing to DRAM via CXL, claiming it can shrink 10-server workloads to one and cut hyperscaler costs.

Stepfun open-sources Step 3.7 Flash, a 196B MoE model delivering 400 tokens/s with native tool calling and multimodal support, built specifically for reliable agent workflows.

Anthropic Funding and Product Updates

• 🔥 Anthropic Series H Raise: Anthropic raised $65B in Series H funding at a $965B post-money valuation,...

Anthropic's Mythos was the "critical triggering factor" for IBM and Red Hat's $5 billion cybersecurity investment to patch open-source...

Two new releases signal stronger options for on-device AI:

• Gemma-4 26B praised for daily coding and experimentation on Mac Studio hardware
  -...

Two new open-source projects target the biggest blockers for production AI agents: accuracy on real data and emergent risks in multi-agent setups.

-...

Qwen-VLA extends the Qwen vision-language stack into continuous action generation via a DiT decoder, unifying manipulation, navigation, and trajectory...

No significant updates today.

No significant updates today.

• OrcaSheets Data Lake offers a secure endpoint for data ingestion with no warehouse setup required.
• Users query unlimited data in plain English,...

Open-source platforms are closing the gap between local AI experiments and production agents.

• Hugging Face analysis: ClementDelangue is...

Two new open-source models target Claude's dominance in AI coding with strong capabilities and lower barriers.

• HY3 Preview delivers real coding...

Major Model Releases

• 🔥 Grok V9 1.5T: xAI completed training of Grok V9 at 1.5 trillion parameters with public release planned in two to three...