Anthropic Unveils Claude Opus 5 with Major Gains in Coding and Knowledge Work
모델 출시 | Tue Jul 28 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 4 sources
Claude Opus 5 launched at half the price of Fable 5 while achieving frontier-level performance, along with benchmark results.
Analysis
[Anthropic] released Claude Opus 5 [1][2][3]
- Half the price compared to Fable 5
- New default model for Claude Max
- Maintains the same $5/$25 per M tokens pricing as Opus 4.8
[Claude Opus 5] achieved SOTA on coding and knowledge work benchmarks [1]
- More than 2x performance improvement over Opus 4.8 on Frontier-Bench v0.1
- 3x the score of the next-best model on ARC-AGI 3
- Best performance on the OSWorld 2.0 computer use benchmark
[Anthropic] introduced cybersecurity safeguards and Automatic Fallbacks [2][3]
- Stronger cybersecurity safeguards compared to Opus 4.8
- Automatic routing to a lower-tier model when the safety classifier triggers
- 85% reduction in classifier trigger frequency compared to Fable 5
[SlopCodeBench] exposed Opus 5's limitations on long-horizon coding tasks [4]
- Checkpoint-based evolutionary coding benchmark
- Opus 5 scored 24% strict pass
- 5x increase in function count in code complexity compared to Opus 4.8