Claude 3 Opus Review: The New Best AI Model?
Anthropic just dropped Claude 3. Three models: Haiku (fast), Sonnet (balanced), and Opus (powerful).
Opus claims to beat GPT-4 on benchmarks.
We spent a week testing it. Here’s what we found.
The Headlines
- 200K context window — That’s ~150,000 words
- Multimodal — Analyzes images and documents
- Near-instant Haiku — Fastest response times we’ve seen
- Benchmark leader — Beats GPT-4 on most tests
What We Tested
Writing Quality
Claude’s always been strong here. Opus is better.
The prose is more natural. Less “AI voice.” Better at matching styles when asked.
For long-form content, the 200K context means it maintains coherence across entire documents. No more losing the plot.
Verdict: Best writing AI we’ve used.
Reasoning and Analysis
We gave Opus complex problems:
- Business case analyses
- Technical architecture reviews
- Logic puzzles
- Code reviews
It handled ambiguity better than GPT-4. More nuanced answers. Better at saying “I don’t know” when appropriate.
Verdict: Matches or beats GPT-4.
Coding
Mixed results. For Python and JavaScript, it’s excellent. For niche languages and frameworks, GPT-4 still has an edge.
Opus explains its code better. The debugging suggestions are clearer.
Verdict: Competitive but not definitively better.
Image Analysis
New capability for Claude. We tested:
- Chart interpretation
- Document analysis
- Photo description
- Diagram understanding
Solid performance. Not as strong as GPT-4 Vision yet, but functional and improving.
Verdict: Good enough for most use cases.
The Pricing
| Model | Price per 1M tokens (input/output) |
|---|---|
| Haiku | $0.25 / $1.25 |
| Sonnet | $3 / $15 |
| Opus | $15 / $75 |
Claude Pro ($20/month) gets you all three models with usage limits.
Opus is expensive via API. Haiku is remarkably cheap.
Who Should Use What
Haiku
- Quick tasks
- High-volume applications
- When speed matters most
- Budget-conscious projects
Sonnet
- Daily use
- Balance of quality and cost
- Most professional tasks
- Default choice for most people
Opus
- Complex analysis
- Long documents
- When quality matters most
- Hard problems
Versus GPT-4
| Category | Claude 3 Opus | GPT-4 |
|---|---|---|
| Writing | Winner | — |
| Reasoning | Tie | Tie |
| Coding | — | Slight edge |
| Images | — | Winner |
| Context | Winner (200K) | — (128K) |
| Speed | Winner | — |
No clear overall winner. Use both for different tasks.
What We Love
- Writing quality — The best we’ve experienced
- Context window — 200K changes what’s possible
- Speed — Especially Haiku
- Nuanced responses — Better at complexity
- Honest about limitations — Says “I don’t know” appropriately
What We Don’t
- No code execution — GPT-4’s Code Interpreter is missed
- Image generation — Not supported
- Plugin ecosystem — ChatGPT has more integrations
- Opus pricing — Expensive for API heavy use
Our Verdict
Claude 3 Opus is legitimately impressive. For writing and analysis, it’s our new default.
But it’s not a clear “switch from GPT-4” situation. The tools are complementary.
Our recommendation:
- Get Claude Pro ($20) for writing and analysis
- Keep ChatGPT Plus ($20) for code and integrations
- Use both based on task
Yes, that’s $40/month. If you use AI professionally, it’s worth it.
The AI landscape just got more competitive. That’s good for all of us.