Claude 3: Anthropic's Latest AI Marvel Outshines Rivals

Calculating... Comments

Anthropic, that up-and-coming AI whiz kid gunning for OpenAI, just dropped their latest creation: Claude 3. Remember when they first rolled out Claude last year? Well, since then, they've been climbing the ranks like a monkey on a banana binge, rubbing shoulders with the bigwigs like Google, Microsoft, and OpenAI.

Anthropic's got some serious backing, too. They’ve got Amazon and Google throwing money their way. Founded by a bunch of former ChatGPT execs, they’re really making waves. Claude 3 isn’t just another pretty face in the crowd—it’s packing some serious heat. Unlike other models that get tangled up in controversies over image generation, Claude 3 can analyze visual content like photos, charts, and graphs without all the drama.

And get this—Claude 3 Opus, their top-of-the-line version, is strutting around beating OpenAI's ChatGPT-4 and Google's Gemini Ultra in some benchmark multiple-choice tests. Pretty impressive, huh? It's like watching the new kid on the block show up and start outshining the neighborhood legends.

claude 3 comparison table Claude 3 Opus beats Gemini and GPT-4 in key benchmarks

Anthropic’s really cranking it up to eleven with Claude 3 Opus. They’re saying this bad boy is setting a whole new standard, blowing the doors off those AI evaluation tests. Apparently, it’s got almost human-like understanding and articulation in some pretty complex tasks.

Looks like Claude 3 Opus is putting them right at the cutting edge of AI innovation. Anthropic’s definitely staking their claim in the AI big leagues.

Opus, our most intelligent model, achieves near-human comprehension capabilities. It can deftly handle open-ended prompts and tackle complex tasks.
— Anthropic (@AnthropicAI) March 4, 2024

Comparison of Claude 3 Models

Model	Intelligence Level	Speed	Cost (Input $/million tokens)	Cost (Output $/million tokens)
Haiku	Fastest	Near-instant	$0.25	$1.25
Sonnet	Balanced	2x faster than Claude 2 and 2.1	$3	$15
Opus	Most intelligent	Similar speed to Claude 2 and 2.1	$15	$75

Availability

Opus and Sonnet are available now for use in claude.ai and the Claude API in 159 countries.
Haiku will be available soon.
Sonnet powers the free experience on claude.ai, while Opus is available for Claude Pro subscribers.

User Feedback So Far

While researching Claude's rival, Mistral, on their Discord server, I've actually come across people mentioning Opus. Naturally, everybody is trying to compare all the existing models. The following comments are, of course, somebody's personal impressions:

Building Responsible AI

According to Anthropic, tey’ve got teams dedicated to keeping a lid on risks like misinformation, CSAM, election interference, biological misuse, and those sci-fi nightmares of AIs going rogue. They’ve even rolled out something called Constitutional AI to make things safer and more transparent.

They’ve fine-tuned their models to address privacy concerns, especially in areas like biology, cyber stuff, and autonomy. They’ve tagged their models as AI Safety Level 2 (ASL-2) under the Responsible Scaling Policy, which means they’re playing it safe. Red team evaluations have shown negligible catastrophic risks so far, which lines up with White House commitments and the 2023 US Executive Order. They’re not just resting on their laurels either—continuous monitoring is in place to keep an eye on things as they edge closer to ASL-3 standards.

Sources: https://www.anthropic.com/news/claude-3-family