The race for the most capable AI
For two and a half years, the most-watched question in AI has been which company has built the most capable model. The answer has shifted multiple times — from OpenAI's runaway lead in 2023, to a multi-way race with Anthropic, Google, Meta, and xAI in 2024–25. Whoever holds the lead in any given quarter sets the prices, the standards, and the direction every other AI company has to react to.
Timeline
- March 14, 2023
OpenAI launches GPT-4. The gap to every other AI company is wide enough that for most of the year, GPT-4 has no real competitor.
- February 15, 2024
Google launches Gemini 1.5 Pro with a one-million-token context window, a breakthrough in long-context understanding.
- March 4, 2024
Anthropic launches the Claude 3 family (Opus, Sonnet, Haiku), introducing three models at different price points. Claude 3 Opus outperforms GPT-4 and Gemini Ultra on several benchmarks.
- April 18, 2024
Meta releases Llama 3, the most capable openly available LLM at the time, establishing Meta as a leader in open-source AI.
- May 13, 2024
OpenAI launches GPT-4o, a multimodal model that handles text, speech, and video in real time at lower cost than previous GPT-4 releases.
- June 20, 2024
Anthropic ships Claude 3.5 Sonnet. In internal agentic coding evaluations, it solves 64% of problems compared to Claude 3 Opus's 38%, marking the first time a non-OpenAI model is widely regarded as better at coding.
- December 11, 2024
Google announces Gemini 2.0 Flash Experimental with native real-time audio and video interactions, signaling Google's push into multimodal AI.
- February 24, 2025
Anthropic launches Claude 3.7 Sonnet, the industry's first hybrid reasoning model that can give both real-time answers and extended step-by-step thinking.
- December 17, 2025
Google launches Gemini 3 Flash and makes it the default model in the Gemini app, positioning it as a fast, low-cost competitor to OpenAI's offerings.
Where things stand right now
The capability race has become genuinely multi-way. OpenAI, Anthropic, Google, Meta, and xAI all have models competitive on different dimensions — coding, reasoning, multimodal, cost, and open availability. No single company holds a clear across-the-board lead, and the advantage shifts quarterly depending on which capability buyers prioritize.