List prices did not move. Tokens fell about a sixth. The independent Index bill still edged up. Cost per finished task is the number that matters.
Via implicator.ai: Claude Opus 5 Cost $3,835 to Benchmark, More Than Opus 4.8
Anthropic released Claude Opus 5 on July 24, 2026 at unchanged API rates of USD 5 per million input tokens and USD 25 per million output tokens - the same list prices as Opus 4.8. implicator.ai reports that Artificial Analysis spent USD 3,835.51 to run Opus 5 on its Intelligence Index under the Adaptive Reasoning, Max Effort configuration, against USD 3,752.55 for Opus 4.8 - roughly a 2.2 percent increase.
Opus 5 generated about 100 million tokens on the suite versus 120 million for its predecessor - about a sixth fewer - yet still landed a higher evaluation bill. It scored 61 on the Index, first among 187 models, ahead of Fable 5 at 60, GPT-5.6 Sol at 59, and Opus 4.8 at 56.
What happened: an independent lab published evaluation spend that rose even as token count fell. Why an SME owner should care: list price and token counts are not the same as cost per finished task in your workflows.
Vendor launch charts often sell lower cost per task. The Artificial Analysis Index run is a different measurement: same configuration, published totals, and a small upward move in dollars despite fewer tokens generated.
implicator.ai notes both Opus 5 and Opus 4.8 were described as very verbose beside the lab's cross-model average of 63 million tokens. That is the quiet trap for SMEs: models that talk more - or think longer - can erase a list-price win before anyone opens the invoice.
Anthropic is pitching an effort setting customers can lower to conserve tokens, and early-access quotes in the same materials claim meaningful token or turn reductions on specific workloads. Harvey's applied research lead cited 26 percent fewer tokens on average versus Opus 4.8 at max reasoning; Fundamental Research Lab reported higher accuracy with fewer turns. Those are workload-specific claims, not a blank check for every firm.
Fable 5 cost about USD 6,200 on the same Index in June, which the lab called its most expensive model benchmarked. Opus 5's run cost roughly 38 percent less while scoring one point higher - useful context, still not your production bill.
implicator.ai also notes Anthropic separately claimed half the cost per task versus Fable 5 on CursorBench 3.2 - a different evaluation from the Intelligence Index. Mixing those charts is how teams talk themselves into a migration without measuring their own work.
Sonnet 5 context from the same publication trail matters too: a June 30 report calculated Sonnet 5 at USD 2.29 per Intelligence Index task, about 15 percent more than Opus 4.8, with the rise attributed entirely to token use at unchanged Sonnet list prices. Effort settings and verbosity are the real knobs.
Treat the Index numbers as a signal to re-baseline, not as a purchase order:
Owner-led firms with 10 to 50 people do not need 187-model leaderboards. They need a short vendor map, a measured unit economics sheet, and someone who owns the switch.
Write the acceptance test before the bake-off. If a research memo must cite sources, a contract redline must preserve defined terms, or a client email must stay under a word count, score those outcomes first. Then record dollars and minutes. Only then compare Opus 5 to whatever you run today.
Artificial Analysis reported USD 3,835.51 to evaluate Opus 5 on its Intelligence Index under Adaptive Reasoning, Max Effort.
AgentsROI is stack-agnostic Managed AI Services for SMEs. We run, govern, and measure AI so it keeps paying for itself - without pushing a single model brand.
Primary fit here is Model Selection and Continuity Planning: match the model to the job on cost, capability, and privacy, with a fallback when a release changes behavior or pricing dynamics.
Supporting fit is a Workflow ROI Audit - find where AI saves real time and money, and where a higher Index score still fails your unit economics. That paid diagnostic is the gate before a retainer.
If you already have tools in production, Managed AI Operations keeps monitoring, optimization, and governance on so a model swap does not become an unowned cost spike.
Opus 5 may be the right upgrade for your stack. The Artificial Analysis bill is a reminder that token cuts and list prices can still leave you paying more for the work that matters.
Book a Model Selection review when you want a vendor-neutral map of what to run, what to fall back to, and what to measure before legal finishes the memo.
Credit: Facts summarized from implicator.ai coverage of Artificial Analysis Intelligence Index measurements and Anthropic Opus 5 launch pricing as of July 24-25, 2026.
This article summarizes publicly reported information and is for general informational purposes only. It does not constitute legal, tax, financial, investment, security, or compliance advice. AgentsROI.ai is not a law firm, accounting firm, or registered investment adviser. Facts, pricing, statistics, and product capabilities cited here reflect the sources listed at the time of writing and may change. Readers should verify current information independently and consult qualified professionals regarding obligations specific to their industry, jurisdiction, and circumstances—including applicable New York State and New York City requirements. AgentsROI.ai may have commercial relationships with vendors mentioned; where material, such relationships are disclosed. Nothing in this article is an endorsement of any specific AI product, model, or provider.
