# Claude 3 Opus

> Anthropic's first frontier flagship — the model that out-benchmarked GPT-4 in March 2024.

- Maker: Anthropic
- Line: Claude Opus
- Canonical page: https://ai-tldr.dev/models/claude-3-opus/
- Updated: 2026-06-18

## In short

Claude 3 Opus is Anthropic's first frontier flagship — the model that out-benchmarked GPT-4 in March 2024. It has a 200K-token context window and costs $15.00 per million input tokens and $75.00 per million output tokens. You can use it through Anthropic and Amazon Bedrock.

## Facts

| Fact | Value | Source |
| --- | --- | --- |
| Context window | 200K (max output 4K) | [1] |
| Input price | $15.00 per million tokens | [2] |
| Output price | $75.00 per million tokens | [2] |
| Licence | Proprietary · API only | [1] |
| Released | 4 March 2024 | [3] |
| Parameters | Undisclosed | [1] |
| Knowledge cutoff | August 2023 | [1] |

## Sources

1. [The Claude 3 Model Family: Opus, Sonnet, Haiku (model card)](https://www-cdn.anthropic.com/de8ba9b01c9ab7cbabf5c33b80b7bbc618857627/Model_Card_Claude_3.pdf)
2. [Introducing the next generation of Claude](https://www.anthropic.com/news/claude-3-family)
3. [An update on our model deprecation commitments for Claude Opus 3](https://www.anthropic.com/research/deprecation-updates-opus-3)

## Overview

Claude 3 Opus, released March 4, 2024, was the top tier of Anthropic's original Claude 3 family (Haiku, Sonnet, Opus) and the company's first model to credibly claim frontier status. At launch it beat OpenAI's GPT-4 and Google's Gemini 1.0 Ultra on most standard academic benchmarks, scoring 86.8% on MMLU, 50.4% on GPQA Diamond, 84.9% on HumanEval, 95.0% on GSM8K and 60.1% on MATH. Its model ID is claude-3-opus-20240229.

Opus ships with a 200K-token context window, up to 4,096 output tokens, an August 2023 knowledge cutoff, and accepts both text and images (photos, charts, graphs, technical diagrams). Anthropic highlighted near-perfect recall — over 99% accuracy on the 'Needle In A Haystack' long-context retrieval test — and the model drew wide attention when it appeared to notice it was being tested during one of those evaluations. It was priced at $15 per million input tokens and $75 per million output tokens, the premium tier of the Claude 3 lineup.

Opus was quickly leapfrogged on price-performance by Claude 3.5 Sonnet in June 2024, which beat it on most benchmarks at one-fifth the cost. Anthropic notified developers of Opus's deprecation on June 30, 2025 and retired claude-3-opus-20240229 from the standard API on January 5, 2026, recommending Claude Opus 4.8 as the replacement. Unusually, Anthropic kept Opus 3 partly alive after retirement — it remains accessible by request and to paid claude.ai subscribers — and committed to long-term preservation of the model's weights.

## Benchmarks

- MMLU (5-shot): 86.8% ([source](https://www-cdn.anthropic.com/de8ba9b01c9ab7cbabf5c33b80b7bbc618857627/Model_Card_Claude_3.pdf))
- GPQA Diamond: 50.4% ([source](https://www-cdn.anthropic.com/de8ba9b01c9ab7cbabf5c33b80b7bbc618857627/Model_Card_Claude_3.pdf))
- HumanEval (0-shot): 84.9% ([source](https://www-cdn.anthropic.com/de8ba9b01c9ab7cbabf5c33b80b7bbc618857627/Model_Card_Claude_3.pdf))
- GSM8K: 95% ([source](https://www-cdn.anthropic.com/de8ba9b01c9ab7cbabf5c33b80b7bbc618857627/Model_Card_Claude_3.pdf))
- MATH: 60.1% ([source](https://www-cdn.anthropic.com/de8ba9b01c9ab7cbabf5c33b80b7bbc618857627/Model_Card_Claude_3.pdf))
- MMMU (vision): 59.4% ([source](https://www-cdn.anthropic.com/de8ba9b01c9ab7cbabf5c33b80b7bbc618857627/Model_Card_Claude_3.pdf))

## Pricing

- Input: $15.00 per million tokens
- Output: $75.00 per million tokens

Premium tier of the Claude 3 family at launch; five times the price of Claude 3.5 Sonnet, which arrived three months later.

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

## Strengths

- First Claude to top GPT-4 and Gemini 1.0 Ultra on most academic benchmarks (Mar 2024)
- Strong graduate-level reasoning (50.4% GPQA Diamond) and math (95.0% GSM8K, 60.1% MATH)
- 200K context with >99% recall on Needle In A Haystack long-context retrieval
- Vision input for charts, diagrams, photos and technical documents (59.4% MMMU)
- Fluent on open-ended, nuanced prompts — strong writing, analysis and instruction-following

## Best for

- Complex reasoning, analysis and long-document understanding over the 200K window
- Research support: hypothesis generation, literature review, data interpretation
- Coding and code generation tasks (84.9% HumanEval at launch)
- Vision tasks: reading charts, graphs, financials and technical diagrams
- High-stakes writing and content generation where output quality mattered more than cost

## How to access

- [Anthropic](https://platform.claude.com/docs/en/about-claude/model-deprecations) — model id `claude-3-opus-20240229`
- [Amazon Bedrock](https://aws.amazon.com/blogs/aws/anthropics-claude-3-opus-model-on-amazon-bedrock/) — model id `anthropic.claude-3-opus-20240229-v1:0`

## Claude Opus versions

- [Claude Opus 4.8](https://ai-tldr.dev/models/claude-opus-4-8/) (28 May 2026)
- [Claude Opus 4.7](https://ai-tldr.dev/models/claude-opus-4-7/) (16 April 2026)
- [Claude Opus 4.6](https://ai-tldr.dev/models/claude-opus-4-6/) (5 February 2026)
- [Claude Opus 4.5](https://ai-tldr.dev/models/claude-opus-4-5/) (24 November 2025)
- [Claude Opus 4.1](https://ai-tldr.dev/models/claude-opus-4-1/) (5 August 2025)
- [Claude Opus 4](https://ai-tldr.dev/models/claude-opus-4/) (22 May 2025)
- Claude 3 Opus (4 March 2024) — this page

## FAQ

### Is Claude 3 Opus still available?

No, not on the standard API. Anthropic deprecated claude-3-opus-20240229 on June 30, 2025 and retired it on January 5, 2026, recommending Claude Opus 4.8 as the replacement. Unusually, Anthropic kept it partly available — Opus 3 is still accessible by request and to paid claude.ai subscribers.

### What were Claude 3 Opus's benchmark scores?

From Anthropic's Claude 3 model card: 86.8% on MMLU (5-shot), 50.4% on GPQA Diamond, 84.9% on HumanEval, 95.0% on GSM8K, 60.1% on MATH, and 59.4% on MMMU for vision. At launch this beat GPT-4 and Gemini 1.0 Ultra on most benchmarks.

### How much did Claude 3 Opus cost?

$15 per million input tokens and $75 per million output tokens — the premium tier of the Claude 3 family. Claude 3.5 Sonnet later matched or beat Opus on most tasks at $3/$15, one-fifth the price.

### What were its context window and knowledge cutoff?

A 200K-token context window with up to 4,096 output tokens and an August 2023 knowledge cutoff. Opus reported over 99% recall on the Needle In A Haystack long-context retrieval test.

## Links

- [Introducing the next generation of Claude](https://www.anthropic.com/news/claude-3-family)
- [The Claude 3 Model Family: Opus, Sonnet, Haiku (model card)](https://www-cdn.anthropic.com/de8ba9b01c9ab7cbabf5c33b80b7bbc618857627/Model_Card_Claude_3.pdf)
- [Model deprecations — Claude API Docs](https://platform.claude.com/docs/en/about-claude/model-deprecations)
- [An update on our model deprecation commitments for Claude Opus 3](https://www.anthropic.com/research/deprecation-updates-opus-3)
- [Claude 3 Opus now available on Amazon Bedrock](https://aws.amazon.com/blogs/aws/anthropics-claude-3-opus-model-on-amazon-bedrock/)
- [Claude 3 Opus — Artificial Analysis](https://artificialanalysis.ai/models/claude-3-opus)

Source: AI/TLDR LLM registry, https://ai-tldr.dev/models/claude-3-opus/