Anthropic Introduces Claude Opus 5: New Model Nearly Matches Fable 5 Level, But Costs Half as Much
"Discover Claude Opus 5 from Anthropic — a new AI model that's nearly reached the level of Fable 5, but at half the price. Read our review!"
Краткий пересказ от QRazy ИИ
- Anthropic has introduced Claude Opus 5 – a powerful AI model for programming and scientific research.
- The new model shows improved results in benchmarks, surpassing previous developments of the company and its competitors.
- Key enhancements in long-term task execution capabilities make Opus 5 an ideal solution for complex automation.
Anthropic has released a new flagship artificial intelligence model, Claude Opus 5, aimed at complex programming, autonomous AI agents, and tasks requiring deep analysis. The company claims that the new system has capabilities comparable to the more powerful Claude Fable 5 while maintaining a significantly lower cost.
A New Model for Complex Tasks and Autonomy
The market for advanced AI models is becoming increasingly competitive: Anthropic is simultaneously facing the development of solutions from OpenAI and the rapid growth of Chinese developments. Against this backdrop, the company has introduced Claude Opus 5—a model it positions as a new level for professional coding, scientific research, and complex intellectual tasks.
Claude Opus 5 replaces Opus 4.8 and retains the previous API access cost. At the same time, this model has become the main offering for Claude Max subscribers and the most powerful version available to Claude Pro users.
According to Anthropic, in a number of popular tests, Opus 5 has managed to outperform not only the company's previous models but also competing solutions, including Fable 5 and GPT-5.6 Sol.
Test Results Show Significant Capability Gains
Anthropic has published the results of several benchmarks that should demonstrate the advantages of the new model in real-world usage scenarios.
- Frontier-Bench v0.1: Claude Opus 5 achieved a score of 43.3%, setting a new record among tested models. This is more than double the score of Opus 4.8, while the cost of task execution has decreased.
- CursorBench 3.2: at maximum computational effort, Opus 5 lagged behind the peak result of Fable 5 by only 0.5%, although its usage costs about half as much.
- ARC-AGI 3: in the test for solving new, previously unknown tasks, the model showed a result about three times higher than its nearest competitor.
- Zapier AutomationBench: Opus 5 demonstrated about 1.5 times more successful task completions compared to the next best model at the same cost.
- OSWorld 2.0: the new model managed to outperform Fable 5 while spending just over a third of its cost.
| Opus 5 | Fable 5 | Opus 4.8 | GPT-5.6 Sol | |
|---|---|---|---|---|
| Agentic terminal coding Frontier-Bench v0.1 | 43.3% | 33.7% | 21.1% | 34.4% |
| Knowledge work GDPval-AA v2 | 1861 | 1747 | 1593 | 1736 |
| Novel problem-solving ARC-AGI-3 | 30.2% | — | 1.5% | 7.8% |
| Agentic search BrowseComp | 90.8% | 87.4% | 84.3% | 90.4% |
| Multidisciplinary reasoning Humanity’s Last Exam | 56.3% no tools 64.7% with tools | 56.5% no tools 63.9% with tools | 49.8% no tools 57.9% with tools | — |
| Computer use OSWorld 2.0 | 70.6% | 66.1% | 55.7% | 62.6% |
| Business workflows AutomationBench | 26.0% | 17.4% | 17.0% | 18.1% |
| Biology BioMistryBench | 49.4% hard 90.1% human solved | 46.5% hard 89.0% human solved | 42.4% hard 88.5% human solved | — |
The Model Has Improved Its Performance on Long Tasks
Anthropic separately notes improvements in scenarios where the AI must carry out a sequence of actions for an extended period. According to the company, Claude Opus 5 better checks its own work, seeks real reasons for errors, rather than just correcting external manifestations of problems.
The model also more frequently returns to previously completed steps and refines the result until the task is genuinely resolved. This is especially important for programming, process automation, and working on large projects, where one answer is insufficient.
At the same time, Anthropic acknowledges that Opus 5 is not a leader in every area. In some tasks related to offensive cybersecurity and specific research in biology, the model still lags behind Mythos 5.
Developers Have More Freedom with Tool Usage
Alongside the release of the new model, Anthropic has also introduced another change for developers. Now they can change the set of tools available to Claude within the current dialogue without the need to reset or recreate the prompt cache.
This simplifies the creation of complex agent systems where the set of available features may change depending on the current task.
The Cost of Claude Opus 5 Remains Unchanged
Claude Opus 5 is already available in all Anthropic products and through the API named claude-opus-5. The cost of using the model is $5 per million input tokens and $25 per million output tokens.
For users who prioritize speed, the company has also launched a Fast mode. This provides about 2.5 times faster request processing but costs twice as much as the standard rate.
The release of Claude Opus 5 indicates that the race for leadership among AI models is increasingly shifting not only towards maximum power but also towards efficiency. The ability to achieve results on par with a more expensive model at a lower cost is becoming one of the main competitive factors among artificial intelligence developers.