Anthropic announced Claude Opus 5, its newest AI model, on July 24, 20261. Pricing stays at $5 per million input tokens and $25 per million output tokens — unchanged from the previous Opus 4.812. On top of that, the company claims results that match or beat its top-tier Claude Fable 5 on several benchmarks13. The model is available from day one across Claude.ai, the Claude API, Claude Code, and Claude Cowork1.
This is the “same price, better performance” kind of model update: for existing Opus users, it is a drop-in upgrade candidate at no extra cost. It also positions Opus 5 as a challenger to Fable 5 at less than half the cost, which shifts the baseline for model selection.
Benchmarks: closing in on Fable 5, ahead in places
According to Anthropic’s published results, Opus 5 surpasses all other models on the hard-problem benchmark Frontier-Bench v0.1, more than doubling Opus 4.8’s score1. On ARC-AGI 3, an abstract-reasoning benchmark, its score is three times as high as the next-best model1, and on Zapier AutomationBench its pass rate is around 1.5 times the next-best model1.
The comparisons against Anthropic’s own top tier are notable. On the coding benchmark CursorBench 3.2, Opus 5 lands within 0.5% of Fable 5’s peak score at half the cost per task1, and on OSWorld 2.0, a computer-use benchmark, it surpasses Fable 5’s best result at just over a third of the cost1. TechCrunch also notes that Opus 5 actually outperforms Fable 5 on a number of benchmarks3.
These are vendor-reported figures, and real-world results will vary by workload. Still, building the announcement around comparisons with a higher-priced model makes the intent clear: bringing Fable 5-class performance down to a mainstream price point.
Pricing and Fast mode: same base price, 2x for 2.5x speed
Standard pricing is $5 input / $25 output per million tokens, identical to Opus 4.812. For reference, the higher-tier Fable 5 costs $10 input / $50 output2, making Opus 5 exactly half price.
A new option is Fast mode, which runs at around 2.5 times the default speed for twice the base price1. For agentic workloads or interactive development where latency is the bottleneck, you can now buy speed without changing models. Since API usage is billed per token, the practical approach is to mix standard and Fast mode by use case.
API specs: 1M-token context window and the effort parameter
Per the official documentation, the model ID is claude-opus-5, with a 1-million-token context window and a 128k-token maximum output2. The training data cutoff is May 20262.
Opus 5 supports the effort parameter for controlling reasoning depth, which defaults to high on the Claude API and Claude Code2. Beyond the Claude API, the model is offered through Amazon Bedrock, Google Cloud, and Microsoft Foundry2. Meanwhile, the older Opus 4.1 will be retired on August 5, 2026, with Anthropic directing users to migrate to Opus 52.
Safety positioning: the “most aligned model to date”
Anthropic describes Opus 5 as its “most aligned model to date” in automated behavioral audits1. At the same time, the company says Opus 5 does not advance the frontier in risky, dual-use capabilities, and remains behind Mythos 5 in both biology research and offensive cybersecurity1.
According to TechCrunch, the safety classifiers are expected to engage 85% less often for Opus 5 than for Fable 5, making Opus both cheaper and less restrictive than Fable in practice3. The division of roles is now clearer: Fable/Mythos pair top capability with strict controls, while Opus targets everyday work with fewer restrictions.
The backstory: the Claude 5 lineup is now complete
Anthropic introduced the new Fable 5 and Mythos 5 tier in June 2026 (Fable 5 reached general availability on June 92) and followed with the mid-tier Sonnet 5 on June 30. Opus 5 completes the main tiers of the Claude 5 generation. On the competitive front, OpenAI made its GPT-5.6 family generally available on July 9, and the pace of flagship-class model updates has only accelerated this summer.
On the product side, Anthropic also announced upgraded Claude voice mode with higher-tier models on July 23, keeping both the model lineup and the user-facing experience moving.
What it means for you: the model-selection baseline just moved
For readers using Claude at work, the implication is simple: Opus 4.8 users should plan to move to Opus 5 at the same cost. For API integrations, pricing is unchanged, so the only migration questions are performance and behavioral compatibility.
If you have been paying for Fable 5, it is worth testing whether Opus 5 is enough for your workloads. On vendor-reported numbers, it reaches comparable levels on coding and computer use at less than half the cost — a sensible path is to run your main workloads side by side and consolidate on Opus 5 where quality holds. That said, these benchmarks are self-reported, and the final call should rest on evaluations against your own tasks.
Sources
- Introducing Claude Opus 5 - Anthropic official announcement (July 24, 2026)
- Models overview - Anthropic official documentation
- Anthropic launches Opus 5 - TechCrunch (July 24, 2026)