Module 1 of 5 · Intermediate · 8 min read
The model landscape in late 2026
Who builds what: frontier labs, model families and the tiers inside each.
- 1Anthropic, OpenAI and Google trade the overall lead; xAI, Meta and Chinese labs are close behind on many races.
- 2Every family has tiers: a flagship, a balanced workhorse and a fast, cheap option.
- 3Open-weights models from DeepSeek, Moonshot, Alibaba and Zhipu trail the frontier by months, not years.
01The frontier labs
Anthropic (Claude), OpenAI (GPT), Google DeepMind (Gemini) and xAI (Grok) release the most capable closed models. Each has a recognisable personality: Claude is known for coding, agentic reliability and writing; GPT for breadth, computer use and creative writing; Gemini for multimodality, long context and research; Grok for speed and real-time information.
Meta, DeepSeek, Moonshot, Alibaba, Zhipu and Mistral push open or partly open models, often at a fraction of the price.
02Tiers within a family
Labs ship several sizes. Knowing the tier tells you more than the version number.
- Flagship: maximum capability, highest price (Claude Opus 5.5, GPT-6 Astra, Gemini 4 Argon).
- Workhorse: most of the capability at a fraction of the cost (Claude Sonnet 5.5, GPT-6.1 Sol).
- Fast: cheap and quick for high volume (Claude Haiku 5.5, GPT-6 Luna, Gemini 3.5 Flash).
03How fast it changes
Leadership changes hands every few weeks in some categories. That is why AIModelRace tracks weeks at #1 and full history, rather than a single snapshot. A model that leads by a photo-finish margin today may be second next month; one with a dominant lead usually stays ahead longer.
Finished reading?
Mark it complete to track your progress, then test yourself with the Models & Labs quiz.