Anthropic subscriptions offer roughly five times OpenAI’s API-equivalent value, research firm SemiAnalysis said after changes to OpenAI’s $200 monthly plan effectively halved its value.. SemiAnalysis estimates Claude subscriptions consume about 42 percent of Anthropic’s inference compute while generating just 10 percent of revenue.
It argues that subscription pricing is therefore important to evaluating AI companies financially.. Subscriptions remain the main purchasing method for individuals and small businesses. Heavy subsidies provide more tokens than metered API pricing, supporting underground markets in many regions for reselling tokens from pooled accounts..
SemiAnalysis compared GPT-6.1 Sol with Claude Opus 5.5. It acknowledged that OpenAI’s model costs much less per token but said Anthropic’s advantage remains substantial when measured by raw token counts.. According to SemiAnalysis, Claude Pro’s $20 monthly plan provides 2.9 billion tokens per month, compared with 1 billion for ChatGPT’s $20 plan..
OpenAI did not immediately respond to a request for comment.. SemiAnalysis measures token-output rates, converts them into totals per five-hour usage window and per month, then calculates API-equivalent value. Its researchers consider that the best overall subscription metric, while allowing that raw token volume can sometimes be more appropriate..
Artificial Analysis instead measures estimated cost per completed task. At API rates, it puts GPT-6.1 Sol at $0.72 per task and Claude Opus 5.5 at $5.98.. Those comparisons have limits: benchmarks may not match actual workloads, and models can use different token counts to produce similarly functional output, including code.
API-based comparisons can also lag capability changes as new open-source models alter what a given price buys.. Businesses cannot count on subscription subsidies and may lack AI cost-management experience. McKinsey reported in July that 93 percent of enterprises exceeded their AI budgets..
Dhruv Amin, co-founder of AI-agent company Anything, told The Register that businesses were seeing rising AI costs. Since the year began, better performance on certain business problems had encouraged wider deployments and rapidly increasing token bills, he said.. Amin described subsidized prosumer plans as a marketing funnel that familiarizes users with tools they then bring to work.
Larger companies have too many employees for subsidized plans, while costs rise quickly at scale, he said.. Enterprises often develop workloads on frontier models before switching to open-source models they control and can run more cheaply, Amin said.. Anything reduced internal AI costs by more than 75 percent by moving workloads to its Skydive Glide model router, Amin said.
The router seeks equivalent open-source models without reducing quality; Anything now runs many workloads on GLM-5.3 Flash.. Amin said effective cost managers maintain workload-specific evaluation benchmarks and use them to test newly released models in their AI harnesses for consistent performance.
He expects companies to delegate more evaluation work to service providers.. When customers deploy an agent on Anything’s platform, the company captures data on its production performance and automatically constructs an evaluation dataset, Amin said.. Featured image credit.
Tags: AnthropicopenAI
