Ep. 011 - GPT 5.5 vs Claude 4.7: OpenAI's Comeback From the Brink (Tokenomics)
Speakers
Max Kan
— Substack author, SemiAnalysis
Dylan Patel
— Founder, CEO, and Chief Analyst at SemiAnalysis
Jordan Nanos
— Member of Technical Staff at SemiAnalysis
Summary
The panel discusses the latest AI model releases including OpenAI's GPT 5.5, Anthropic's Claude 4.7, and DeepSeek's delayed v4. They analyze model quality, token cost, benchmark limitations, and China's compute constraints. The consensus is that while OpenAI has regained competitiveness, the long-term advantage may shift to companies with larger compute resources like Meta.
- GPT 5.5 brings OpenAI back to frontier but not definitively ahead of Claude.
- Claude 4.7's tokenizer change increases token costs by 35%.
- DeepSeek's release shows China hitting compute constraints.
- Benchmarks are seen as increasingly unrepresentative of real-world use.
- Fast mode vs quality tradeoff becomes more relevant as model costs rise.
- Meta is expected to pull ahead of Chinese AI due to compute scale.
- SubQ model claims are dismissed as likely fake.
- GPU price increases signal continued compute demand.
Trade Ideas
Meta's compute advantage will widen gap.
Meta will pull away from Chinese AI companies like DeepSeek and Kimi because they have a massive compute advantage, having signed huge deals and rebuilt their AI team, and the amount of compute is a key input to model quality. This gap will become evident in the second half of 2026 or first half of 2027.
Up Next
This SemiAnalysis Weekly video, published May 06, 2026, features Max Kan discussing META. 1 trade idea extracted by AI with direction and confidence scoring.