Anthropic has launched Claude Opus 5, priced at $5 per million input tokens and $25 per million output tokens, matching the cost of Opus 4.8 while delivering substantial performance improvements, particularly in agentic work. The model is now the default on Claude Max and available through Claude Pro.
TL;DR
- Anthropic releases Claude Opus 5 at the same price as Opus 4.8, claiming it approaches Fable 5 performance at half the cost.
- Opus 5 shows significant gains in agentic work, scoring 43.3% on Frontier-Bench v0.1 compared to Fable 5's 33.7% and Opus 4.8's 21.1%.
- The model also performs well in knowledge work and computer use, but lags behind Fable 5 and GPT-5.6 Sol in some benchmarks.
What happened
Anthropic has introduced Claude Opus 5 across all platforms, maintaining the pricing of its predecessor, Opus 4.8, at $5 per million input tokens and $25 per million output tokens. The new model is designed to offer substantial improvements in coding, knowledge work, computer use, and scientific research. Developers can access it through the Claude API using claude-opus-5, and it is now the default model on Claude Max and the strongest model available through Claude Pro.
Anthropic positions Opus 5 as an everyday alternative to Claude Fable 5, claiming it approaches Fable 5's performance at half the price. However, the company acknowledges that Opus 5 trails Fable 5 on some benchmarks and is not as capable as Mythos 5 in offensive cybersecurity and long-running biological research. Customers can adjust the model's effort setting to balance reasoning, token use, speed, and cost.
Why it matters
Opus 5's performance in agentic work is particularly notable, with a score of 43.3% on Frontier-Bench v0.1, compared to 33.7% for Fable 5 and 21.1% for Opus 4.8. This makes it a strong contender for tasks requiring sequential actions. Additionally, Opus 5's cost efficiency is highlighted by its performance on CursorBench 3.2, where it came within 0.5% of Fable 5's peak result while costing half as much per task.
For developers and startups, Opus 5 offers a more affordable option for integrating advanced AI capabilities into their products. Investors may see this as a strategic move by Anthropic to capture a larger market share by offering high-performance models at competitive prices. However, the model's limitations in certain areas, such as offensive cybersecurity and long-running biological research, should be considered.
Key facts
- Claude Opus 5 is priced at $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8.
- Opus 5 scores 43.3% on Frontier-Bench v0.1, compared to 33.7% for Fable 5 and 21.1% for Opus 4.8.
- The model achieves a 26% pass rate on AutomationBench, higher than Fable 5's 17.4% and GPT-5.6 Sol's 18.1%.
- Opus 5 scores 1,861 on GDPval-AA v2, outperforming Fable 5's 1,747, Opus 4.8's 1,593, and GPT-5.6 Sol's 1,736.
- The model shows a 10.2 percentage point improvement over Opus 4.8 in inferring molecular structures from spectroscopy data.
- Opus 5 passes 79.4% of vulnerability-identification challenges on OSS-Fuzz, compared to 80% for Mythos 5 and 61.5% for Opus 4.8.
- Anthropic's behavioral audit gives Opus 5 an overall misaligned-behavior score of 2.3, the lowest among its recent models.
- Opus 5 can be run in fast mode at around 2.5 times its standard speed, costing twice the model's base price on the Claude Platform.
Context
Anthropic's release of Claude Opus 5 comes at a time when the AI industry is increasingly focused on cost efficiency and performance optimization. The model's ability to approach the performance of more capable models like Fable 5 at a lower cost is a significant development, as it makes advanced AI capabilities more accessible to a broader range of users. This move also highlights Anthropic's strategy to offer a diverse portfolio of models tailored to different use cases and budget constraints.
The AI landscape is rapidly evolving, with companies continuously pushing the boundaries of what AI models can achieve. Anthropic's focus on agentic work and its commitment to improving performance in scientific research and knowledge work position Opus 5 as a strong contender in the market. However, the model's limitations in certain areas, such as offensive cybersecurity and long-running biological research, underscore the ongoing challenges in developing AI models that excel across all domains.
