Claude 3: Has Anthropic Finally Beaten ChatGPT?
Claude 3 launched March 4, 2024, with Anthropic claiming to beat GPT-4 for the first time. Three models: Opus, Sonnet, and Haiku.
On March 4, 2024, Anthropic released Claude 3—and made a bold claim: Claude 3 Opus beat GPT-4 on major benchmarks.
This was huge. Since GPT-4's release a year earlier, nothing had consistently beaten it. Claude 3 Opus did.
The Three Models
Anthropic released a family:
Claude 3 Haiku: Fast, affordable, for simple tasks Claude 3 Sonnet: Balanced performance and speed Claude 3 Opus: Most capable, beats GPT-4
The tiered approach let users choose based on needs and budget.
The Benchmark Results
Claude 3 Opus outperformed GPT-4 on:
- MMLU (general knowledge)
- Graduate-level reasoning
- Math problem-solving
- Coding challenges
- Multilingual tasks
For the first time, OpenAI wasn't the clear leader.
New Capabilities
Vision: All three models could process images 200K context: Maintained from Claude 2.1 Reduced refusals: Less likely to refuse valid requests Better instruction following: More accurately did what you asked
Where Are They Now?
Claude 3 proved Anthropic was a legitimate GPT-4 competitor. The competition drove both companies to improve faster.
March 4, 2024 was the day OpenAI's model dominance ended—the AI race became truly competitive.