Model Details
Claude Opus 4.8
Quick answer: Claude Opus 4.8 is the latest versioned checkpoint in Anthropic's Opus 4 series as of mid-2026, succeeding Opus 4.7. It maintains the Opus 4 line's focus on deep reasoning, complex agentic workflows, and sustained long-horizon task performance at $15/$75 per million tokens.
What Claude Opus 4.8 Is
Claude Opus 4.8 is Anthropic's most recent Opus 4 checkpoint update, continuing the series that began with the flagship Opus 4 release in May 2025. The Opus 4 series has followed a consistent trajectory: each numbered update refines tool-use reliability, improves instruction adherence in complex agent workflows, and tightens alignment properties. The series reached 81.42% on SWE-bench Verified at Opus 4.6 — subsequent checkpoints build on that foundation.
For teams that have built production systems on the Opus 4 series, the numbered update path provides incremental improvements without requiring architecture changes.
Specifications
| Field | Value |
|---|---|
| Organization | Anthropic |
| License | Proprietary |
| Modality | Multimodal (text and vision) |
Pricing
| Input (per 1M tokens) | Output (per 1M tokens) | |
|---|---|---|
| Anthropic | $15.00 | $75.00 |
Related Models & Benchmarks
- Compare: Claude Opus 4.7 · Claude Opus 4.6
- Benchmarks: SWE-bench Verified
Last updated 2026-06-19.
Claude Opus 4.8 vs Alternatives
| Model | ARC-AGI | Agents Last Exam | LHTB | License |
|---|---|---|---|---|
| Claude Opus 4.8 | 92.5% | 45.1% | 49.2% | Proprietary |
| Claude Opus 4.7 | 93.5% | 41.8% | — | Proprietary |
| Claude Opus 4.5 | 37.6% | — | — | Proprietary |
Opus 4.8 vs 4.7: nearly equal ARC-AGI (92.5% vs 93.5%) but higher Agents Last Exam (45.1% vs 41.8%). 4.8 appears slightly better at agentic tasks.
Frequently Asked Questions
What is Claude Opus 4.8?
Anthropic's May 2026 model scoring 92.5% ARC-AGI, 45.1% Agents Last Exam. Proprietary — competitive with Opus 4.7.Scores from Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.