claude-opus-5
Meant to excel at
No sourced claims yet.
Watch out for
-
Would not compare exact model names and described itself as an older Opus in a routing prompt; prompt by capability rather than model name.
Practitioner report i10x.ai
Benchmarks
Coding percentile 0.74
| Benchmark | Score | Percentile | Source | Date |
|---|---|---|---|---|
| lmarena code arena webdev | 1689.3 | 0.74 | Independent evaluator lmarena.ai | undated |
Agentic coding percentile 0.66
| Benchmark | Score | Percentile | Source | Date |
|---|---|---|---|---|
| swe-bench pro | 79.2 | 0.83 | Vendor claim, relayed datanorth.ai | 2026-07-27 |
| swe-bench verified | 96 | 1.00 | Vendor claim, relayed datanorth.ai | 2026-07-27 |
| terminal-bench 2.1 (aggregator, anotherwrapper.com) | 84.6 | 0.25 | Independent evaluator anotherwrapper.com | 2026-10-19 |
| terminal-bench 4.0 | 52.3 | 0.56 | Vendor's own site anthropic.com | 2026-09-22 |
Overall percentile 1.00
| Benchmark | Score | Percentile | Source | Date |
|---|---|---|---|---|
| artificial analysis intelligence index | 61 | 1.00 | Independent evaluator datanorth.ai | 2026-07-27 |
Percentile is the share of tracked models this model beats on that benchmark, averaged across the group. Only benchmarks at least four tracked models share are scored.
Practitioner reports
- -
In Endor Labs' 179-task run Opus 5 was the slowest and most expensive of three Claude models (about 14 min per task, 15 timeouts, roughly $1,116 total), versus 4 min and $116 for Opus 5.5.
Task: bug-fix | Compared with claude-opus-5.5 | 2026-09-24
Practitioner report endorlabs.com - -
Every's team says Opus 5 ignored skills and instructions that Opus 5.5 follows, and its prose was about a reading grade harder.
Task: writing-content | Compared with claude-opus-5.5 | 2026-09-22
Practitioner report every.to - +/-
On a single small bug (averaging an empty list), Opus 5 and GPT-5.5 both found the ZeroDivisionError and proposed a guard, scoring a tie; Opus also suggested raising ValueError. One run per prompt, editorial scoring, and the publisher sells a multi-model workspace.
Task: bug-fix | Compared with gpt-5.5 | 2026-08-21
Practitioner report i10x.ai