# Claude Opus 5 Launches Near Fable 5's Power at Half the Price

Page: https://hirantha.me/blog/claude-opus-5-launches-near-fable-5s-power-at-half-the-price
Published: 2026-07-31
Written by: hirantha.me Newsroom (AI-generated). An AI model wrote this from the sources listed below; how it is made: https://hirantha.me/about/newsroom
Topics: anthropic, claude-opus-5, llm-benchmarks, ai-pricing, model-release

> Anthropic's Claude Opus 5 nearly matches its flagship Fable 5 on benchmarks while costing about half as much to run.

## A cheaper model that mostly keeps up with the flagship

Anthropic's newest model, Claude Opus 5, became available on 24 July 2026, and its headline feature is not raw capability but price. The company is positioning it as landing close to the intelligence of its top-tier Claude Fable 5 model while costing roughly half as much to run - a break from the usual pattern where the strongest model is also the most expensive one.

Opus 5 is priced at five dollars per million input tokens and twenty five dollars per million output tokens (a token being the small chunk of text a language model reads or writes at a time), which happens to match what the previous Opus 4.8 cost. Cached prompts are cheaper still: writing to the cache costs $6.25 per million tokens for a five-minute cache lifetime, and reading from a cache hit costs $0.50 per million tokens. Anthropic also offers a "fast mode" at double the base price that runs about 2.5 times quicker. Opus 5 is now the default model for subscribers on the Claude Max plan and the strongest model available to Claude Pro subscribers, and it is reachable through the API under the model name claude-opus-5.

## Where it actually lands against the competition

Independent benchmarking outlet Artificial Analysis scored Opus 5 at 61 on its Intelligence Index, ahead of Fable 5 (60), OpenAI's GPT-5.6 Sol (59), Kimi K3 (57), and the older Opus 4.8 (56). On the Artificial Analysis Coding Index, Opus 5 ties GPT-5.6 Sol for first place, and it scored 89 percent on Terminal-Bench v2.1, a benchmark of command-line task completion, when run at its highest "max" reasoning setting.

The gaps get more interesting on individual tests. On a knowledge-work benchmark called AA-Briefcase, Opus 5 reached an Elo rating (a relative skill score borrowed from chess ranking systems) of 1720 at maximum reasoning effort, 146 points clear of Fable 5, while costing between $10.41 and $17.79 to run depending on reasoning tier against $22.30 for Fable 5 - up to a 53 percent saving for a stronger result. On Anthropic's own Frontier-Bench v0.1, the company says Opus 5 roughly doubles Opus 4.8's score while beating every competing model on the list. On CursorBench 3.2, a coding-assistant benchmark, it lands within half a percentage point of Fable 5's score for half the price. On ARC-AGI 3, a test of abstract reasoning, Anthropic says it scores roughly triple the next-best model, and on the computer-use benchmark OSWorld 2.0 it beats Fable 5 at about a third of the cost.

It is not a clean sweep. Opus 5 scored 53 percent on Humanity's Last Exam, tying Fable 5 but trailing rivals on the CritPt physics benchmark. More notably, its hallucination rate, meaning how often it states something false with confidence, rose to 50 percent, fourteen points higher than Opus 4.8.

## New controls, and a narrower safety gap

Anthropic added an "effort" toggle that lets a developer choose low, medium, or high effort to trade capability and speed against cost, alongside a new beta feature called Automatic Fallbacks that reroutes a request to a less capable model instead of returning an error when it trips a safety trigger. Unlike Fable 5 and Anthropic's Mythos 5 model, Opus 5 does not use a 30-day data retention policy.

Anthropic describes Opus 5 as its most aligned model to date, with the lowest rate of deceptive behavior of any Claude release, and says its enhanced cybersecurity safety classifiers now intervene roughly 85 percent less often than they did for Fable 5. It still trails Anthropic's own Mythos 5 specifically on cybersecurity exploitation tasks. As one example of its reasoning ability, Anthropic said the model wrote its own computer vision pipeline in response to a prompt that left out key details, and the company also highlighted particular strength in scientific research tasks such as organic chemistry and protein analysis, along with improved visual output.

Opus 5 is notable simply for its pace of arrival: it is the fourth Claude model Anthropic has shipped in under two months, following Mythos 5, Fable 5, and Sonnet 5, and it lands roughly two months after Opus 4.8, which launched on 28 May 2026.

## What it means in practice

For developers already building on Claude, Opus 5 is a straightforward upgrade path: similar or lower cost than Opus 4.8 for a meaningfully stronger model, plus a reasoning-effort dial that makes it easier to control spend on high-volume workloads. The rising hallucination rate is the one figure worth pausing on - for tasks where factual precision cannot be traded away, pairing Opus 5's output with a verification step or a retrieval-grounded workflow is a reasonable precaution rather than an overreaction, and it's a reminder that benchmark leadership and reliability are not the same axis.

## Sources

- [Introducing Claude Opus 5](https://www.anthropic.com/news/claude-opus-5), Anthropic
- [Anthropic launches Opus 5](https://techcrunch.com/2026/07/24/anthropic-launches-opus-5/), TechCrunch
- [Anthropic's Claude Opus 5 costs well below Fable 5 while matching or beating it across most benchmarks](https://the-decoder.com/anthropics-claude-opus-5-costs-well-below-fable-5-while-matching-or-beating-it-across-most-benchmarks/), The Decoder