AI News

ClaudeDevs publishes a guide for building with Claude Sonnet 5.5

ClaudeDevs says Claude Sonnet 5.5 is out and has published a guide for building with it, covering model choice against Opus 5.5, migration from Sonnet 5, and use in Claude Code.

Databricks makes Lakebase Search generally available in Postgres

Databricks says Lakebase Search is now generally available, adding scalable vector search and BM25 full-text search inside Postgres, with pay-per-use pricing and no compute cost while idle.

Polymarket says Walmart vows not to use AI for individualized pricing

Polymarket says Walmart has vowed it will never use AI to analyze customers’ income or shopping history in order to charge different prices for the same products.

Opus 5.5 leads Anthropic models on OpenRouter by spend and tokens

About one week after launch, OpenRouter says Opus 5.5 ranks first among Anthropic models on its platform by share of spend and share of tokens, with a particularly rapid shift away from Opus 5.

Claude can now help build evaluations and hillclimb on them

ClaudeDevs says Claude can help build evaluations and hillclimb on them, and has published guidance on eval design and skills that Claude Code can use to improve applications.

Arena.ai ranks GPT-6 Luna (Max) 23rd in Agent Arena

Arena.ai reports a +1.6% net improvement for OpenAI’s GPT-6 Luna (Max) across 8,000 real-world agentic sessions, at a $0.05 median cost per task. It did not reach the Agent Arena Pareto frontier.

Sonnet 5.5 is now available in Cursor

Cursor says Sonnet 5.5 is now available and performs on par with Opus in many tasks.

AMD agrees to acquire World Labs

AMD said it has a definitive agreement to acquire World Labs and will bring the company’s researchers and model experts into AMD to work on future AI infrastructure and the open AI ecosystem.

Team Bots enter public beta for Teams and Enterprise

Team Bots are now in public beta for Teams and Enterprise. Grok Bot says SpaceXAI uses them for account preparation, engineering coordination, data questions, customer feedback, and hiring.

Grok Bot introduces Team Bots as shared AI teammates

Team Bots are shared AI teammates that learn as a team works with them. Each bot can be given the skills, plugins, and credentials for its role and used in Slack or Grok Bot.

Melissa Pan says coding-agent harness choice affects cost more than accuracy

Melissa Pan compared Claude Code, Codex, and Pi and reports that the system around an AI model, the harness, affects cost more than accuracy. She cites three findings; only this one is stated in the announcement.

shimrit ben-yair says Google AI Pro unlocks a 24/7 personal AI agent in Gemini

shimrit ben-yair says Google AI Pro can now unlock a 24/7 personal AI agent in Gemini to take on routine tasks. More information is on the Google AI plans page.

Polymarket says OpenAI teases 20+ launches at DevDay 2026

Polymarket says OpenAI has teased more than 20 launches for OpenAI DevDay 2026, describing the event as happening tomorrow.

Cline says Sonnet 5.5 beats Opus 5.5 on Terminal-Bench 4.0 at half the cost

Cline says Sonnet 5.5 beats Opus 5.5 on Terminal-Bench 4.0 while costing half as much. Cline also says it uses fewer tokens than earlier models, cuts cost by up to 30% per task, and generates output 30% faster than its predecessor.

Ali Ghodsi says GPT-6 Luna is 20x cheaper than Opus 5.5 for high-volume routing

Ali Ghodsi says a test of the latest AI models on 2,400 engineers shifted the cost/quality frontier. He says GPT-6 Luna is 20 times cheaper than Opus 5.5 for high-volume routing, while Opus 5.5 is now their daily default, with higher quality and a 20% lower cost than prior baselines.