Coship lays out four tiers of AI intelligence

The post, dated October 11, 2026, is a snapshot rather than a permanent ranking. The author notes that a model is in a tier at the point of release, plus or minus a few days, and will likely fall behind afterward. The lab-specific analysis is labeled as of November 10th, 2026, and may be outdated for later readers.

The tier structure rests on access to resources and knowledge. Tier 1 requires top people, high-quality data, and abundant compute, and cannot rely on distillation from competitors. Tier 2 sits behind the frontier but above the best open source, with labs able to use top-tier models for R&D even though Anthropic, OpenAI and Google ban building competitors with their models. Tier 3 gains full access to top open models, including reasoning traces and weights, making it more achievable. Tier 4 is a catch-all for the rest, though the author stresses that doesn't mean models there are bad. The rule is that moving within a tier is easier than crossing between them.