Claude 3 Launch: Everything Anthropic Announced
Anthropic just launched Claude 3. Three models. Major improvements. GPT-4 competition intensifies.
Here’s everything that was announced.
The Three Models
Claude 3 Haiku — Fast
Position: Quick, cheap, efficient Best for: Simple tasks, high-volume applications Speed: Near-instant responses Price: Cheapest of the three
Claude 3 Sonnet — Balanced
Position: The everyday workhorse Best for: Most tasks, general use Balance: Good speed and capability Price: Middle tier
Claude 3 Opus — Powerful
Position: Maximum capability Best for: Complex reasoning, professional use Claims: Matches or beats GPT-4 on benchmarks Price: Premium
Key Improvements
1. Vision (Finally)
Claude can now see images. Analyze photos, read documents, understand charts.
This was missing before. Major addition.
2. 200K Context
Process ~150,000 words in a single conversation. Analyze entire books, large codebases, extensive documents.
Claude’s context window is now larger than GPT-4’s.
3. Better Reasoning
Benchmark scores suggest improved reasoning across math, logic, and analysis tasks.
4. More Honest
Anthropic claims Claude 3 is better at admitting uncertainty and avoiding hallucinations.
5. Faster
Significant speed improvements, especially with Haiku.
Benchmark Claims
| Benchmark | Claude 3 Opus | GPT-4 |
|---|---|---|
| MMLU | 86.8% | 86.4% |
| HumanEval | 84.9% | 67.0% |
| GSM8K | 95.0% | 92.0% |
Benchmarks aren’t everything, but these are competitive numbers.
Pricing
Per Token (API)
| Model | Input (per 1M) | Output (per 1M) |
|---|---|---|
| Haiku | $0.25 | $1.25 |
| Sonnet | $3.00 | $15.00 |
| Opus | $15.00 | $75.00 |
Claude Pro ($20/month)
All three models available. Usage limits apply.
What This Means
For Users
More options. Claude is now competitive with GPT-4 on capabilities while offering different strengths.
For Developers
Three tiers mean you can optimize cost vs. capability. Haiku for simple stuff, Opus for complex.
For the Market
Real competition. This is good for everyone.
Claude 3 vs. GPT-4
| Aspect | Claude 3 Opus | GPT-4 |
|---|---|---|
| Writing | Slight edge | Good |
| Coding | Competitive | Strong |
| Context | 200K | 128K |
| Vision | New | Established |
| Ecosystem | Smaller | Larger |
| Price | Similar | Similar |
Neither is definitively better. Both are excellent.
First Impressions
What We Like
- Vision finally available
- Haiku is genuinely fast
- Writing quality remains excellent
- Context window is massive
What We’re Watching
- Real-world performance vs. benchmarks
- API reliability at scale
- How Claude Pro usage limits work
Availability
- Claude.ai: All users (free and Pro)
- API: Available now
- Enterprise: Contact Anthropic
Our Take
Claude 3 is a legitimate major release. The gap between Claude and ChatGPT has closed significantly.
For users: You now have two excellent choices. For OpenAI: Real pressure to respond.
Competition is good. Users win.
Major Claude release. We’ll have detailed reviews after more testing.