Model Details
Claude Opus 4
Quick answer: Claude Opus 4 is Anthropic's May 2025 flagship model, scoring 72.5% SWE-Bench Verified, 46.9% LiveCodeBench, 6.68% HLE, and 8.6% ARC-AGI v2. Priced at $15/$75 per MTok.
At a Glance
Where Claude Opus 4 leads
- 72.5% SWE-Bench Verified — competitive real-world software engineering
- Anthropic brand trust and safety research investment
- Strong long-context reasoning
- Extended thinking mode
Where it lags
- 6.68% HLE — lower than newer frontier models
- 8.6% ARC-AGI v2 — moderate on abstract reasoning
- $15/$75 per MTok — high cost vs Claude Sonnet 4
- Succeeded by Claude models in subsequent generations
Best for: Enterprise use cases requiring Anthropic's safety guarantees; complex agentic software engineering; long-context document analysis.
What Claude Opus 4 Is
Claude Opus 4 is Anthropic's May 2025 flagship — the most capable model in the Claude 4 generation at release. Claude models are known for their adherence to Constitutional AI principles, strong instruction following, and careful reasoning on complex tasks.
The 72.5% SWE-Bench Verified score reflects Opus 4's strength on real-world software engineering tasks — competitive at the time of release. The model supports extended thinking mode for longer chain-of-thought reasoning on complex tasks.
At $15/$75 per MTok, Opus 4 is positioned at the premium tier of Claude 4, with Claude Sonnet 4 and Haiku 4 offering lower-cost alternatives.
Specifications
| Field | Value |
|---|---|
| Organization | Anthropic |
| License | Proprietary (API only) |
| Release date | May 22, 2025 |
| Modality | Text only |
| Context window | 200K tokens |
| Output tokens | Up to 32K |
Pricing
| Tier | Price per MTok |
|---|---|
| Input | $15.00 |
| Output | $75.00 |
Available via Anthropic API (api.anthropic.com) and Amazon Bedrock.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| SWE-Bench Verified | 72.5% | Benchgen evaluation | 2025-05 |
| LiveCodeBench | 46.9% | Benchgen evaluation | 2025-05 |
| Humanity's Last Exam | 6.68% | Benchgen evaluation | 2025-05 |
| ARC-AGI v2 | 8.6% | Benchgen evaluation | 2025-05 |
| Shade Arena | 30.2% | Benchgen evaluation | 2025-05 |
Claude Opus 4 vs Alternatives
| Model | SWE-Bench | LiveCodeBench | Price (in/out) |
|---|---|---|---|
| Claude Opus 4 | 72.5% | 46.9% | $15/$75 |
| GPT-5 Codex | 74.5% | — | Proprietary |
| HY3 | 78.0% | — | Proprietary |
| DeepSeek-V3.2 Speciale | 73.1% | — | MIT (open) |
Claude Opus 4 at 72.5% SWE-Bench is competitive with GPT-5 Codex (74.5%) and HY3 (78%). For open-source SWE: DeepSeek-V3.2 Speciale (73.1%, MIT). For Anthropic ecosystem: Opus 4 with extended thinking is the primary choice.
Related Models & Benchmarks
- Compare: Claude Sonnet 3.7 · GPT-5 Codex · DeepSeek-V3.2 Speciale
- Benchmarks: SWE-Bench Verified · LiveCodeBench · Humanity's Last Exam
Frequently Asked Questions
What is Claude Opus 4?
Anthropic's May 2025 flagship model scoring 72.5% SWE-Bench Verified, 46.9% LiveCodeBench. $15/$75 per MTok. Top of the Claude 4 generation.Is Claude Opus 4 worth $15/$75?
For tasks requiring maximum Claude capability (complex reasoning, long-context, extended thinking), Opus 4 is justified. For most production coding/chat tasks, Claude Sonnet 4 offers better price/performance. Use Opus 4 when you need the flagship.Specs from Anthropic's Claude Opus 4 release (May 2025) and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.