Claude Opus 5 Arrives: Anthropic Claims It Matches Fable and Is More Cost-Effective

Anthropic has just raised the stakes in the AI race with Claude Opus 5. After releasing Claude Fable 5 to the public a few weeks ago, Anthropic claims that Opus 5 offers similar capabilities but at half the price.

In other words, Opus 5 can be used to achieve performance comparable to Claude Fable 5—an AI model whose use was temporarily banned in the US due to its immense power—in tasks such as programming and knowledge-based work, but at a lower cost.

The launch of Claude Opus 5 comes at a time when US AI labs are facing pressure from Chinese open-weight models, such as Kimi K3 and GLM 5.2, to deliver high performance at affordable prices.

How good is Claude Opus 5?

To put the assessment of Fable 5 into context: the United States initially banned foreigners from using that AI model for cybersecurity reasons. Now, Opus 5 offers similar performance but is geared toward everyday use.

Anthropic states that Opus 5 "offers greatly improved performance for the same cost as its predecessor, Opus 4.8," allowing users to adjust effort settings to prioritize intelligence or save tokens, thereby obtaining faster and more cost-effective results.

According to benchmarks shared by Anthropic, Claude Opus 5 not only matches but outperforms Fable 5 in 10 of the 14 metrics analysed, approaching Fable 5's frontier-level intelligence at a lower operating cost. Opus 5 also managed to outperform GPT-5.6 Sol—OpenAI’s flagship model—in selected tests.

Regarding performance benchmarks in software and knowledge-based work, Anthropic noted that Opus 5 has set new standards. In Frontier-Bench v0.1, it outperformed competing models and more than doubled the performance of Opus 4.8, all at a lower cost per task. In CursorBench 3.2, it came within 0.5% of Fable 5’s top score but at half the cost.

Anthropic cited early examples from the testing phase. These included instances where Opus 5 wrote its own computer vision pipeline to reconstruct a mechanical part as a 3D model in FreeCAD, and helped an engineer at a trading firm create a market data feed for a new trading platform in a single session.

Opus 5 is not trained for cybersecurity

While Claude Opus 5 can perform at the level of Fable in certain tasks, Anthropic has clarified that it comes with major safeguards when it comes to cybersecurity. The AI startup says that Opus 5 was not state-of-the-art for such “risky, dual-use capabilities,” and remains behind Mythos 5 in offensive cyber and biology research.

Following the cybersecurity concerns arising from Claude Mythos, Anthropic says that it had intentionally avoided training Opus 5 on cyber tasks. While Opus 5 can allow developers to “identify and fix software vulnerabilities,” it will block “high-risk uses.”

Such safeguards likely allowed Anthropic to face less pushback from the US over the release of the model.

Claude Opus 5 is cheaper than Fable

Claude Opus is priced at $5 (roughly Rs 480) per million input tokens and $25 (roughly Rs 2,410) per million output tokens, matching Opus 4.8. For context, Claude Fable 5 is priced at $10 (roughly Rs 960) per million input tokens and $50 (roughly Rs 4650) per million output tokens.

This is particularly important as enterprises are starting to rethink AI usage due to mounting costs. Be it Uber or Walmart, many companies are restricting how employees use AI models.

Though Opus 5 remains a bit more expensive than Chinese open-weight AI models. For instance, Moonshot AI’s Kimi K3 is cheaper, priced at $3 (roughly Rs 290) for every million input tokens, and $15 (roughly Rs 1,450) for every million output tokens.

In case you are unaware, tokens act as a basic unit of measurement for AI use. The more work you do, the more tokens you use.

Claude Opus 5 is now available as the default model for Claude Max. It is also the strongest available model for Claude Pro users.