Claude Opus 4.8 vs Claude Opus 5
| Spec | Claude Opus 4.8 | Claude Opus 5 |
|---|---|---|
| Respan Index | 55.7 (#27) | 67.0 (#16) |
| Blended price per 1M | $10 | $10 |
| Context window | 1M | 1M |
| Input / output per 1M | $5 / $25 | $5 / $25 |
| Max output | 128K | 128K |
| Input | text, vision | text, vision |
| Released | 2026-05-28 | 2026-07-24 |
Respan Index by category
Each category is scored 0 to 100, and how far ahead the winner is. Leads under a point are close to even.
| Category | Claude Opus 4.8 | Claude Opus 5 |
|---|---|---|
| Coding | 39.4 Claude Opus 4.8 | 64.0 Claude Opus 5 |
| Agents | 64.4 Claude Opus 4.8 | 70.4 Claude Opus 5 |
| Reasoning | 64.8 Claude Opus 4.8 | 74.0 Claude Opus 5 |
| Knowledge & language | 52.5 Claude Opus 4.8 | 61.1 Claude Opus 5 |
| Human preference | 55.2 Claude Opus 4.8 | 58.5 Claude Opus 5 |
Benchmarks both report
Claude Opus 5 leads on 24 of the 30 benchmarks both report. Higher is better. Each score links to where it was published.
| Benchmark | Claude Opus 4.8 | Claude Opus 5 |
|---|---|---|
| LMArena Elo | 1482 | 1490 |
| LMArena Vision (LMArena) | 1286 | 1285 |
| LMArena WebDev (LMArena) | 1555 | 1692 |
| MirrorCode (Epoch AI) | 65.6% | 87.8% |
| Chess Puzzles (Epoch AI) | 52% | 43% |
| LMArena Agent (LMArena) | 0.0776 | 0.0626 |
| APEX-Agents (Mercor) | 48.9% | 65.8% |
| DeepSWE v1.1 (Datacurve) | 59% | 73.6% |
| EBR-bench (Epoch AI) | 28.6% | 45.7% |
| FrontierMath Tier 4 v2 (Epoch AI) | 56.1% | 73.2% |
| FrontierMath Tiers 1-3 v2 (Epoch AI) | 80% | 85.6% |
| Furniture Assembly (Epoch AI) | 42.5% | 60.8% |
| GPQA Diamond | 93.6% | 93.9% |
| LiveBench Agentic Coding (LiveBench) | 50.5% | 65.2% |
| LiveBench Coding (LiveBench) | 81.8% | 81.4% |
| LiveBench Data Analysis (LiveBench) | 66% | 74.6% |
| LiveBench Instruction Following (LiveBench) | 72% | 63.8% |
| LiveBench Language (LiveBench) | 79.7% | 88.7% |
| LiveBench Mathematics (LiveBench) | 94.3% | 95.7% |
| LiveBench Reasoning (LiveBench) | 89.2% | 91.2% |
| Mystery Game Puzzles (Epoch AI) | 36% | 59% |
| OTIS Mock AIME 2024-2025 (Epoch AI) | 98.3% | 98.9% |
| SAGE (Vals AI) | 54.8% | 49.4% |
| SimpleQA Verified | 53% | 59.9% |
| Terminal-Bench 4.0 (Vals AI) | 23.2% | 53.5% |
| Vending-Bench 2 (Andon Labs) | 5787.43 | 11181.87 |
| WeirdML (Håvard Tveit Ihle) | 82.9% | 91.8% |
| tau2-bench Banking Knowledge (Sierra) | 39.7% | 48.7% |
| SimpleBench (SimpleBench) | 64.8% | 80.6% |
| Humanity's Last Exam (no tools) | 49.8% | 56.6% |
FAQ
Is Claude Opus 4.8 better than Claude Opus 5?
Claude Opus 5 scores 67.0 on the Respan Index (#16), ahead of Claude Opus 4.8 at 55.7 (#27). Claude Opus 5 leads on 24 of the 30 benchmarks both report.
Which is cheaper, Claude Opus 4.8 or Claude Opus 5?
Claude Opus 4.8 lists at $5 input and $25 output per 1M tokens; Claude Opus 5 at $5 and $25. Both cost about the same per token at list price.
Which has the larger context window?
Both accept 1M tokens of context.
Compare other models
- vsClaude Opus 4.8 vs Claude Opus 4.7
- vsClaude Opus 5.5 vs Claude Opus 5
- vsGemini 4 Argon vs Claude Opus 5
- vsClaude Opus 4.8 vs NVIDIA Nemotron 3 Ultra 550B-A55B NVFP4
- vsClaude Opus 4.8 vs Muse Spark 1.3
- vsClaude Opus 4.8 vs Step 5 Preview
- vsClaude Opus 4.8 vs MiMo-V2.6-Pro
- vsClaude Opus 5 vs GPT-6.2 Sol
- vsClaude Opus 5 vs Hy4
- vsClaude Opus 5 vs GPT-6.1 Terra