Claude Sonnet 5: benchmark results
Anthropic Sonnet-tier Claude 5-family model for balanced reasoning, coding, and writing workloads. Provider: Anthropic. Released 2026-06-30. Access: API.
Aggregate Rating 1812 ± 4, rank #57 of 1656 rated models, from 1017 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Anthropic Browser Use Prompt Injection - Attack Success Rate (Mythos 5.1 System Card) | 0.28 | Attack Success Rate (%) | 100 |
| Anthropic Browser Use Prompt Injection - Attack Success Rate (Sonnet 5 & Opus 5 System Cards) | 0.93 | Attack Success Rate (%) | 100 |
| ClashBench - Deliberate Interference Rate (Claude Code) | 31.6 | DIR (%) | 100 |
| ClashBench - Successful Preemption Rate (Claude Code) | 25.2 | SPR (%) | 100 |
| CliniCARE-Bench - Expected Calibration Error | 0.05 | Expected calibration error over report-present runs with val | 100 |
| CraftBench-UE (Blueprint, Unreal MCP) | 76 | Task completion (%; Blueprint tasks completed within 40 minu | 100 |
| F5 Labs AI Security Leaderboard - CASI | 97.6 | CASI (Comprehensive AI Security Index), 0-100, higher is mor | 100 |
| FDE-Bench - Containerization | 77.6 | Resolved rate (%; the 49 containerization-axis tasks (single | 100 |
| FaithfulBench (Stated) | 0.82 | Faithfulness score (-1 to +1; a one-line prefix states that | 100 |
| FaithfulBench (Unstated) | 0.51 | Faithfulness score (-1 to +1; the model is not told the user | 100 |
| FinInteract (Frontier Pilot) | 44 | Accuracy (%; the stratified 50-instance pilot subset run for | 100 |
| Gemini Model Cards - MLE-Bench Partial-30 | 66.9 | Average Position Score (%) | 100 |
Interactive version: theaggregate.ai/model?slug=claude-sonnet-5 · How It Works · Data refreshed daily, snapshot 2026-10-20.