AI Task Costs Vary Wildly Across New Models
New benchmarks reveal that output volume, not just token pricing, determines the true cost of running AI agent tasks.
Updated on Oct. 29, 2026 in Artificial Intelligence
Live Poll
Do you trust that AI software providers offer transparent pricing for the tasks you complete?
Novel Cognition has published a comparative analysis of AI task costs following the release of new models from OpenAI and Google. The report highlights how varying token output volumes significantly impact total expenses across popular large language models.
Why it matters
Understanding actual task costs is critical as model providers shift from simple input-output pricing to usage models driven by reasoning intensity. This analysis enables developers to accurately forecast infrastructure spending for AI-driven workflows.
Artificial Analysis measured task costs using its Intelligence Index v5.0 suite, which covers office, coding, and science tasks. While many providers charge $2 per million input tokens and $10 for output, actual task costs vary due to differences in output length.
The players
Novel Cognition
A research entity focused on benchmarking the economic and performance efficiency of large language models.
Artificial Analysis
A platform providing comparative testing and standardized performance metrics for AI inference models.
OpenAI
A developer of large-scale foundation models including the GPT-6.1 series.
A cloud and AI provider that maintains the Gemini series of large language models.
Anthropic
An AI research lab known for its Claude family of large language models.
The details
The analysis uses a nine-test suite to simulate real-world AI agent behavior, including terminal coding and science-based problem solving. Because reasoning models differ in their verbosity, the total bill for an identical request is determined by the volume of output tokens generated. These metrics reveal that a model with identical per-token pricing can result in drastically different overhead depending on its internal chain-of-thought processing depth.
Timeline
26 October 2026: OpenAI released GPT-6.1 Terra.
28 October 2026: Google released Gemini 4 Carbon.
29 October 2026: Novel Cognition performed the pricing analysis.
30 November 2026: Google Gemini 4 Argon price increase begins.
The Tech Race
This analysis arrives as providers rapidly update model families, including recent releases like OpenAI’s GPT-6.1 Terra and Google’s Gemini 4 Carbon. It highlights a shift in the competitive landscape where inference efficiency is now as critical as raw performance in the race for enterprise adoption.
Developers and companies using LLM APIs must now monitor per-task costs rather than just flat token rates, as output verbosity significantly affects monthly bills. Starting 1 December 2026, users of Gemini 4 Argon should update their cost projections to account for the planned doubling in per-token pricing.
The takeaway
Model providers are increasingly differentiating products through complex pricing structures that reward efficiency in output generation. Track the upcoming 1 December 2026 Gemini price shift as an indicator of how AI providers plan to manage profitability for high-inference workloads.
What happens next
Watch for the scheduled price increase for Google Gemini 4 Argon, which rises to $4 per million input tokens and $20 per million output tokens on 1 December 2026.
Further reading
For broader trends in model benchmarking and performance, visit the Artificial Intelligence section.
More information
Review the full AI agent cost analysis report to compare specific model billing metrics.
Source note: This article includes information reported by The Cherry Creek News.
Live Poll
Do you trust that AI software providers offer transparent pricing for the tasks you complete?
