Compare / head-to-head
Claude Opus 4.8vsClaude Sonnet 4.5
Claude Opus 4.8 leads 9 of 9 shared benchmarks. Claude Sonnet 4.5 is 1.7x cheaper per token. Claude Opus 4.8 has the larger context window (1M tokens).
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
9 – 0
Claude Opus 4.8 leads
Cheaper per token
Claude Sonnet 4.5
1.7x cheaper, input + output
Larger context
Claude Opus 4.8
1M tokens
Provider uptime (30d)
100%
Anthropic
Claude Opus 4.8
Anthropic · released 2026-05-28
textvision
Claude Sonnet 4.5
Anthropic · released 2025-09-29
textvision
Quality
Benchmark matrix
9 shared · 7 only Claude Opus 4.8 · 11 only Claude Sonnet 4.5| Benchmark | Claude Opus 4.8 | Claude Sonnet 4.5 | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 9 | ||||
| SWE-bench Verified | 88.6% ↗ | 77.2% ↗ | +11.4 pt | Claude Opus 4.8 |
| GPQA Diamond | 93.6% ↗ | 83.4% ↗ | +10.2 pt | Claude Opus 4.8 |
| Humanity's Last Exam (no tools) | 49.8% ↗ | 17.7% ↗ | +32.1 pt | Claude Opus 4.8 |
| FrontierMath Tier 4 v2 (Epoch AI run) | 56.1% ↗epoch | 2.4% ↗epoch | +53.7 pt | Claude Opus 4.8 |
| FrontierMath Tiers 1-3 v2 (Epoch AI run) | 80% ↗epoch | 23.9% ↗epoch | +56.1 pt | Claude Opus 4.8 |
| Humanity's Last Exam (with tools) | 57.9% ↗ | 33.6% ↗ | +24.3 pt | Claude Opus 4.8 |
| OSWorld-Verified | 83.4% ↗ | 61.4% ↗ | +22 pt | Claude Opus 4.8 |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | 98.3% ↗epoch | 77.8% ↗epoch | +20.5 pt | Claude Opus 4.8 |
| SimpleQA Verified | 53% ↗epoch | 30.7% ↗epoch | +22.3 pt | Claude Opus 4.8 |
| Only Claude Opus 4.8 reports · 7 | ||||
| AIME 2026 | 100% ↗matharena ⚠ | not reported | — | — |
| LMArena Elo | 1452.7 ↗ | not reported | — | — |
| BrowseComp | 84.3% ↗ | not reported | — | — |
| SWE-bench Multilingual | 84.4% ↗ | not reported | — | — |
| SWE-bench Multimodal | 38.4% ↗ | not reported | — | — |
| SWE-Bench Pro | 69.2% ↗ | not reported | — | — |
| Terminal-Bench 2.1 | 74.6% ↗ | not reported | — | — |
| Only Claude Sonnet 4.5 reports · 11 | ||||
| AIME 2025 | not reported | 84.2% ↗matharena ⚠ | — | — |
| ARC-AGI-2 (Verified) | not reported | 13.6% ↗ | — | — |
| GDPval-AA | not reported | 1276 ↗ | — | — |
| MCP Atlas | not reported | 43.8% ↗ | — | — |
| MMMLU | not reported | 89.5% ↗ | — | — |
| MMMU-Pro (no tools) | not reported | 63.4% ↗ | — | — |
| MMMU-Pro (with tools) | not reported | 68.9% ↗ | — | — |
| SWE-bench Verified (Epoch AI run) | not reported | 71.3% ↗epoch | — | — |
| Tau2-bench Retail | not reported | 86.2% ↗ | — | — |
| Tau2-bench Telecom | not reported | 98% ↗ | — | — |
| Terminal-Bench 2.0 | not reported | 51% ↗ | — | — |
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is Claude Opus 4.8 minus Claude Sonnet 4.5 in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing
Side by side
Official pricing pages and model cards| Spec | Claude Opus 4.8 | Claude Sonnet 4.5 | Edge |
|---|---|---|---|
| Context window | 1M tokens | 200K tokens | Claude Opus 4.8 |
| Max output | 128K tokens | 64K tokens | Claude Opus 4.8 |
| Input price / 1M | $5 ↗ | $3 ↗ | Claude Sonnet 4.5 |
| Output price / 1M | $25 ↗ | $15 ↗ | Claude Sonnet 4.5 |
| Cached input / 1M | $0.5 ↗ | $0.3 ↗ | Claude Sonnet 4.5 |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | $30 | $18 | Claude Sonnet 4.5 |
| Modalities | text · vision | text · vision | Tie |
| Released | 2026-05-28 | 2025-09-29 | — |
| Cited benchmark scores | 17 | 21 | — |
Reliability
Anthropic status
Last 60 days · refreshed daily on this page · live on /statusMore matchups
Claude Opus 4.8 vs …
Models sharing the most benchmarksMore matchups
Claude Sonnet 4.5 vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run Claude Opus 4.8 and Claude Sonnet 4.5 on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.