Archive · 25 Jul 2026
Anthropic Releases Opus 5 Globally with Enhanced Performance
Anthropic has released a new AI model called Opus 5.
This model is really good at helping with software engineering tasks.
It performs better than other models and costs less.
For example, it does more than twice as well as the previous model, Opus 4.8, on certain tasks.
It also does almost as well as another top model, Fable 5, but at half the cost.
However, to keep things safe, Anthropic has made sure that Opus 5 is not as advanced as some of their other models when it comes to security checks.
Opus 5 is now the default model for Claude Max users and is also available through the Claude API.
It has a Fast mode that operates around 2.5 times faster than its standard mode.
Anthropic releases Opus 5 globally, excelling in software engineering tasks.
Opus 5 surpasses all other models and more than doubles Opus 4.8’s performance at a lower cost.
On CursorBench 3.2, Opus 5 performs within 0.5% of Fable 5’s peak score at half the cost.
Opus 5 achieves greater performance at a given cost than all other models on high, xhigh, and max effort.
Anthropic has limited Opus 5 behind other models on security evaluations to maintain safeguards.
- Who
- Anthropic
- What
- Release of Opus 5, a new AI model
- When
- July 24
- Where
- Globally
- Why
- To provide enhanced performance and cost efficiency in software engineering tasks
Performance and Cost Efficiency
Security and Safeguards
Performance Metrics
Performance and Cost Efficiency
Opus 5 excels on valuable software engineering tasks, surpassing all other models and more than doubling Opus 4.8’s performance at a lower cost per task.
Security and Safeguards
Opus 5 is intentionally limited behind other models from the firm on security evaluations to maintain safeguards.
- Model Name
- Opus 5
- Performance Benchmark
- Frontier-Bench v0.1
- Performance Improvement
- More than doubles Opus 4.8’s performance
- Cost Efficiency
- Lower cost per task compared to other models
- CursorBench 3.2 Performance
- Within 0.5% of Fable 5’s peak score at half the cost
- Security Evaluations
- Limited behind other models to maintain safeguards
- Release Date
- July 24
- Default Model
- Claude Max users
- Fast Mode Speed
- 2.5 times faster than standard mode
- Pricing
- $5 per million input tokens and $25 per million output tokens
