Eliza Times

Industry News

• Anthropic planning IPO at $380B valuation: Anthropic is reportedly planning to go public this year at a $380B valuation with zero profit, following OpenAI's for-profit transition. Both safety-focused labs are choosing the same capital path. link

• Anthropic gains 11% user boost from Super Bowl ad: Data shows Anthropic received an 11% user increase from its OpenAI-critical Super Bowl commercial, marking a rare win for brand advertising in AI. link

• Kimi K2.5 now available in EU via Nebius: Kimi K2.5 is finally accessible on EU cloud infrastructure through Nebius, expanding availability beyond China. link

Tips & Techniques

• Ask agents "Are you missing any context?": When an agent isn't performing as expected, prompt it with "Are you missing any context?" to surface gaps in its understanding before debugging further. link

• Claude Code can debug live remote servers: Point Claude at a remote box with real data and services to let it refactor code, run migrations, debug logs, or clean up tech debt directly—no local development needed. link

• Use Claude Code for cloud infrastructure audits: Connect Claude Code to your cloud CLIs (AWS, Azure, GCloud) and it can audit real infrastructure, review configs, pull logs, analyze usage, and provide cost projections—replacing expensive consulting engagements. link

• Sierra's approach to LLM failover for high availability: Sierra published insights on balancing consistent agent behavior while failing over to multiple heterogeneous LLM providers to achieve high availability in production. link

New Tools & Releases

• MiniMax M2.5 achieves SOTA on SWE: MiniMax M2.5 reached state-of-the-art on SWE benchmarks through extensive training across 10+ programming languages with significant focus on iOS and Android, leading to major improvements in mobile app development. link

• x-cli tool for Claude Code and other agents: A new CLI tool (not MCP server) that enables Claude Code, Codex, OpenClaw, and other agents to interact with X/Twitter programmatically. link

• Resend disables bot detection for OpenClaw: Resend disabled bot detection on their login flow so OpenClaw agents can now write, send, and manage emails directly—showing how companies are adapting to agent-first workflows. link

Research & Papers

• GPT-5.2 contributes to new theoretical physics result: GPT-5.2 helped derive a new formula for gluon amplitudes in theoretical physics, representing journal-level frontier research where the AI played a crucial role. Physicist Andy Strominger called it the first time AI solved a problem in his field that humans might not have. link

• ChatGPT 5.2 still hallucinates 25% of citations: Despite claims that hallucinations are solved in 2026, testing shows GPT-5.2 hallucinates over 10 of 40 citations—and most users don't enable thinking mode (the default) because AI companies want to save costs. link

• Intelligent AI Delegation framework proposed: New research proposes a framework for when and how AI systems should delegate tasks, addressing key R&D challenges in multi-agent coordination. link

• Agent skills benchmark shows mixed real-world results: 105 domain experts evaluated agent skills across Claude Code, Gemini CLI, and Codex—revealing significant gaps between claimed and actual performance in production environments. link

--- *Curated from 657 tweets across 4 lists*

---

Emerging Trends

🔥 OpenClaw AI Agent Platform (380 mentions) - RISING AI agent platform gaining massive traction with users deploying autonomous agents for tasks like email management, coding, content creation, and business automation. Multiple deployment services emerging.

🔥 Codex 5.3 Adoption (195 mentions) - RISING OpenAI's Codex continuing strong adoption for coding tasks, with users reporting high productivity gains and switching from other tools. CEO Sam Altman expressed surprise at rapid adoption pace.

🔥 Seedance 2.0 Video Generation (168 mentions) - RISING ByteDance's new video generation model causing excitement with impressive quality, recreating viral memes like "Will Smith eating spaghetti" and enabling rapid content creation for creators.

🔥 GLM-5 Model Release (142 mentions) - RISING Major open-weight model release from Zhipu AI with 745B parameters, showing strong performance in coding and agentic tasks. Now available on multiple platforms including Zai.

🔥 MiniMax M2.5 Model Release (98 mentions) - RISING New MiniMax M2.5 model released with strong performance on SWE benchmarks and extensive multi-language training. Users reporting it as competitive alternative to expensive frontier models.