Anthropic's Opus 5 Cut Tokens 17% but Cost More to Benchmark Than Opus 4.8

Anthropic released Claude Opus 5 on July 24 with per-token API rates unchanged from Opus 4.8, the company announced. Artificial Analysis reported that its independent Intelligence Index run cost $3,835.51. The lab published the measurement the same day Anthropic promoted "lower cost per task" and an effort control intended to reduce token use.

The July 24 benchmark cost $3,835.51

Under the same Adaptive Reasoning, Max Effort configuration on the Intelligence Index, Artificial Analysis reported evaluation costs of $3,835.51 for Opus 5 and $3,752.55 for Opus 4.8. Calculated from those listed totals, the increase was roughly 2.2%. The new model generated 100 million tokens during the suite, down about a sixth from its predecessor's 120 million. The lab described both releases as "very verbose" beside its cross-model average of 63 million tokens.

At API rates of $5 per million input tokens and $25 per million output tokens, Opus 5 costs the same per token as Opus 4.8. The new model scored 61 on the Intelligence Index, ranking first among 187 models; Fable 5 scored 60, GPT-5.6 Sol 59 and Opus 4.8 56. Using the two published totals, this article calculates that each Index point cost roughly 6.2% less for Opus 5, giving the newer model more measured performance for each dollar spent. That calculation applies only to this benchmark suite and does not establish the cost of using Opus 5 generally.

Fable 5 cost about $6,200 to run on the same Index in June, which the lab called the most expensive model it had ever benchmarked. The Opus 5 run cost roughly 38% less and scored one point higher; Anthropic separately made a half-cost-per-task claim for CursorBench 3.2, an evaluation different from the Intelligence Index.

Anthropic adds an effort setting

Anthropic's launch charts plot performance against an effort setting that customers can lower to conserve tokens. The company claimed Opus 5 more than doubled Opus 4.8's Frontier-Bench v0.1 performance at a lower cost per task. On CursorBench 3.2, Anthropic reported performance within 0.5% of Fable 5 at maximum effort and half the cost per task.

Early-access customers quoted in the same materials also reported lower resource use. Niko Grupen, Harvey's head of applied research, said the model maintained "similar performance while generating 26% fewer tokens on average compared to Opus 4.8 at max reasoning." Richard Pham, Fundamental Research Lab's evals and product lead, reported 9 percentage points higher accuracy with roughly one-third fewer turns and tool calls and 60% less time across effort levels. Zapier CEO and co-founder Wade Foster said Opus 5 "topped Zapier's AutomationBench leaderboard without spending more tokens than prior Claude models"; he added that previous models did not pass while Opus 5 hit 100%.

Sonnet 5 cost $2.29 per task

In a June 30 report, the lab calculated Sonnet 5 at $2.29 per Intelligence Index task, around 15% more than Opus 4.8. Sonnet 5 kept the $3-per-million input price and $15-per-million output price of Sonnet 4.6, and the lab attributed the per-task increase entirely to token use; the new release generated about 40% more output tokens per task than Sonnet 4.6. A February report found Opus 4.6 used 30% to 60% more tokens than Opus 4.5 on GDPval-AA and was the costliest model the lab had measured there. Both reports tied the higher task bill to token consumption at unchanged prices.

GDPval-AA's 220-task evaluation

The lab's published Opus 5 cost figure covers the Intelligence Index. Its earlier Sonnet 5 and Opus 4.6 reports included per-task findings from GDPval-AA; the Opus 4.6 run used about 160 million tokens to complete 220 tasks. Anthropic says customers can lower Opus 5's effort setting to conserve tokens, making the selected setting one factor in the resulting bill.

Meta Prices Its First Paid AI Model API at $4.25 per Million Output Tokens

Meta Superintelligence Labs released Muse Spark 1.1 on Thursday and opened a public preview of the Meta Model API, the first time the company has charged outside developers for access to a model it bu

The Implicator

Sonnet 5 Closes Most of the Gap to Opus 4.8 on Agent Work

Anthropic released Claude Sonnet 5 on June 30 with a simple pitch: most of Opus 4.8's capability at well under half the cost. On the company's own agent benchmarks, the model trails its flagship by a

The Implicator

OpenAI GPT-5.5 by the Numbers. Higher Prices, Disputed Benchmarks

OpenAI released GPT-5.5 on April 23, 2026, priced at $5 input and $30 output per million tokens on the API, with rollout to ChatGPT Plus, Pro, Business, and Enterprise tiers first while API access fol

The Implicator

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai