How Claude 3 Opus Beat GPT 4 — And Learned to See The Claude Story Ep 4

On 4 March 2024, Anthropic didn't launch one model; it launched three. Claude 3 Haiku, Sonnet and Opus brought vision, a 200K-token context window and frontier-level benchmarks to Claude for the first time. In Episode 4 of The Claude Story, we cover: The Claude 3 family: Haiku (fast and cheap), Sonnet (balanced), Opus (most intelligent) Pricing: $0.25 / $3 / $15 per million input tokens Vision: reading photos, charts, PDFs, handwriting and diagrams Claude 3 Opus vs GPT-4 and Gemini 1.0 Ultra on MMLU, GPQA and GSM8K 200K context window (1M+ for select customers) The needle-in-a-haystack test: the pizza-topping sentence that Opus caught Claude 3 Haiku speed: ~21,000 tokens per second, 400 court cases for about $1 Claude goes global in 159 countries ⏱️ Chapters 0:00 Intro 0:10 The pizza-topping test 0:50 Episode 4: Claude 3 Going Multimodal 1:10 The Claude 3 family: Haiku, Sonnet, Opus 2:20 Claude 3 can see: vision 3:00 Real-world examples 3:10 Claude 3 Opus at the frontier: benchmarks 4:00 200K context window 4:30 Needle-in-a-haystack: 99% recall 5:00 Claude 3 Haiku: speed at a low cost 5:30 Claude goes global: 159 countries 6:00 A smarter tomorrow 6:20 Recap and next episode ▶️ Next: Episode 5 – Claude 3.5 Sonnet, Artifacts and Computer Use 📺 Full playlist – The Claude Story: [PLAYLIST LINK] 🔔 Subscribe for the next chapters of The Claude Story. The Claude Story is an independent product-history documentary series. It is not affiliated with or endorsed by Anthropic. Sources: Anthropic's public announcements. #Claude #Claude3 #AnthropicAI #TheClaudeStory #AI