Some ideas for what comes next, May 2026
Interconnects · Nathan Lambert · 2026-05-26
AI researcher Nathan Lambert argues that open-weight models still trail closed frontier models by at least 12 months on agentic tasks, that Google cannot yet match Claude Code or Codex, and that growing social resistance to AI data centers represents the most underappreciated near-term barrier to continued AI development.
Appears in
Extraction
Topics: open-source-aiagentic-codingfrontier-modelsai-predictionsai-economics
Claims
- Open-weight models have not achieved the agentic coding performance of Anthropic's Opus 4.5, and the gap may persist for 12 or more months.
- Even Google lacks a meaningful competitor to Claude Code and OpenAI Codex for agentic coding workflows.
- An open-weights equivalent to Anthropic's Mythos model is unlikely to appear in 2026 due to resource constraints at Chinese labs and conservative stances at well-funded Chinese corporate labs.
- American open models such as Gemma 4 and Nemotron are gaining adoption and challenging Qwen's previous dominance at comparable parameter sizes.
- AI is bifurcating economic outcomes, rewarding very large companies and very small niche businesses while squeezing mid-tier knowledge workers.
- Social and political resistance to AI data centers is the most underappreciated and underexamined barrier to continued AI progress.
Key quotes
The Opus 4.5 in Claude Code moment of December 2025 was so loud and obvious, that if open models hit this performance level for price points as low as $5/month, there will be an explosion in usage.
AI is beginning to drive companies to the two extremes in the scaling era.
The real position that a large swath of Americans has is that they have a voice in saying no to the current trend — by not granting permission to build data centers.