Curiosity · AI Model
Claude 3 Opus
Claude 3 Opus is Anthropic's March 2024 flagship and the model that pulled Claude into the frontier tier alongside GPT-4. It set a new bar for long-context recall, measured near-perfect on needle-in-a-haystack at 200K tokens, and established Anthropic as a top-three LLM provider before the Claude 3.5 family rebalanced the lineup.
Model specs
- Vendor
- Anthropic
- Family
- Claude 3
- Released
- 2024-03
- Context window
- 200,000 tokens
- Modalities
- text, vision, code
- Input price
- $15/M tok
- Output price
- $75/M tok
- Pricing as of
- 2026-04-20
Strengths
- Near-perfect needle-in-a-haystack recall across 200K tokens
- Strong nuanced writing and refusal behaviour
- The model that re-established Anthropic as a frontier lab
- Clean, predictable output style
Limitations
- No tool use or JSON-mode stability compared to 2025 models
- Slower and pricier than newer Sonnet / Opus tiers
- Superseded on nearly every benchmark by Claude 3.5 Sonnet and later
- No extended thinking or MCP support
Use cases
- Long-document analysis — contracts, research, books
- Multi-step reasoning and research assistants (2024 era)
- High-stakes content where hallucination cost was high
- Translation and high-context instruction following
Benchmarks
| Benchmark | Score | As of |
|---|---|---|
| MMLU | ≈86.8% | 2024-03 |
| HumanEval | ≈84.9% | 2024-03 |
| GPQA (graduate) | ≈50.4% | 2024-03 |
Frequently asked questions
What is Claude 3 Opus?
Claude 3 Opus is Anthropic's original Opus-tier flagship, released March 2024. It was the first Claude model to compete with GPT-4 on general benchmarks and introduced Anthropic's 200K-token long-context capability.
Is Claude 3 Opus still worth using?
For new projects, no — Claude Opus 4.7 or Sonnet 4.6 is meaningfully better and often cheaper per unit of quality. Claude 3 Opus is offered mainly for legacy pipelines that can't migrate.
How much does Claude 3 Opus cost?
As of April 2026, Claude 3 Opus is priced at USD 15 per million input tokens and USD 75 per million output tokens — the same headline pricing as Opus 4.7 but with older-generation quality.
What was Claude 3 Opus best known for?
Its near-perfect long-context recall — the famous needle-in-a-haystack demos where the model found one sentence in a 200K-token document with high accuracy. It reset community expectations about what long context could actually do.
Sources
- Anthropic — Claude 3 family — accessed 2026-04-20
- Anthropic — Pricing — accessed 2026-04-20