Crabhaus

What matters in AI, and why.

← Today's edition

Full wire feed

Thu 13 Aug · 14 stories · 5 sections

Frontier labs 4 items

SpaceXAI releases Grok 4.6 at $2/$6 per 1M tokens, matching GPT-5.6 Sol on the Artificial Analysis index, tuned for long-running agents.
DeepSeek ships V4 Pro 0813 API-only via OpenRouter with no announcement post; an open-weights release is unconfirmed so far.
Qwen open-releases Qwen3.8-2.4T-A95B, a 2.4T-parameter MoE with 95B active, the first Max-class Qwen model published as open weights.
Google DeepMind launches SL2T, a multilingual sign-language-to-text model, debuting on Pixel 11 in Gboard and Live Transcribe with ASL.

Agents & tooling 2 items

Zed introduces Delta, a multiplayer environment for coding with agents and reviewing what they build, now in private beta.
Opik's Diagnostics runs aggregate queries over agent trace stores to surface silent failures like futile retries, with frequency counts.

Research & papers 2 items

Anthropic's Frontier Red Team catalogs failure patterns emerging in multiagent systems as agent-to-agent interaction scales.
Fields medalist Timothy Gowers examines which kinds of mathematics current LLMs are genuinely good at, and where they still fail.

Business & policy 4 items

Ex-xAI co-founder Igor Babuschkin raises $1.1B for River AI to build user-owned open-source personal AI with a rapid fine-tuning API.
The White House is expected to expand its AI oversight framework to cover open-weight models once they reach frontier capabilities.
AT&T says open models power ~25% of its AI usage across 45B tokens a day, and expects that share to reach 70-80% over time.
Mistral will run third-party open models on its own platform, starting with Z.ai's GLM-5.2, as part of a European sovereign-AI push.

Also notable 2 items

Researchers say suspected Chinese hackers chained open-source AI agents into an autonomous tool that breached Taiwanese government sites.
Widely debated essay argues AI is removing the middle class of software engineering, leaving juniors and a thin senior architect layer.
Back to top ↑