Anthropic Claude subscription offers roughly five times the API-equivalent value of OpenAI's plan

A study by SemiAnalysis concludes that Anthropic's subscription plans deliver substantially more value per dollar than OpenAI's, following the latter's recent price adjustments. Industry reporter Thomas Claburn covered the story, noting that OpenAI's $200 per month tier effectively halved its value after changes announced on social media. SemiAnalysis examined subscription pricing across AI companies because individual and small business buyers rely on these tiers as the primary purchase route.

AI subscriptions carry heavy subsidies. Customers receive far more tokens at subscription rates than they would at metered API prices. The research firm argued that understanding subscription pricing matters because subscription usage consumes a disproportionate share of company compute resources.

Anthropic's compute usage versus revenue share

SemiAnalysis reports that Claude subscriptions account for about 42 percent of Anthropic's inference compute while generating only 10 percent of its revenue. Despite this imbalance, the analysis states that Claude subscriptions provide considerable value. The firm claims Anthropic offers approximately five times the API-equivalent value across the board when compared to OpenAI's mid-tier models.

The comparison references GPT-6.1 Sol and Claude Opus 5.5. Even if one argues that 6.1 Sol carries a lower per-token price than Opus 5.5, the gap remains massive, SemiAnalysis says. The firm measured token output rates, converted those into tokens per five-hour usage window and per month, then computed API-equivalent value — its preferred metric for capturing subscription value, though the researchers acknowledge other measures like raw token volume may suit different contexts.

Token volume and cost-per-task comparisons

Artificial Analysis, a separate tracker, measures cost per task rather than token volume. By that yardstick, GPT-6.1 Sol carries an estimated price of $0.72 per task, while Claude Opus 5.5 runs $5.98. The ratio may not hold for every use case, however. Benchmark performance does not guarantee identical results on real workloads, and different tasks may draw varying token counts from different models.

On a per-month basis, Claude Pro at $20 provides 2.9 billion tokens, according to SemiAnalysis. ChatGPT at the same $20 monthly price supplies 1 billion tokens. Model efficiency varies — some jobs may require more tokens from one model and fewer from another, but the functional output can remain comparable.

Enterprise perspective on AI cost management

Dhruv Amin, co-founder of the AI agent business Skydive, told The Register that all businesses are watching AI costs climb. Since the start of the year, better capabilities have driven wider deployment, and token bills have grown accordingly. Labs subsidize their prosumer plans heavily as part of a marketing funnel, Amin said, because early familiarity often leads to workplace adoption.

For larger companies, two problems surface. Subsidized plans cannot accommodate every employee, and costs scale quickly at enterprise volume. Amin described a strategy some teams use: get a workload running well on a frontier model, then switch it to an open source model the company controls and serves at a much lower cost. His own operation moved many workloads to Skydive Glide, a model router that finds equivalent open source models for given tasks without quality drops. Reported results show AI costs dropped by 75 percent or more using this approach.

Amin said companies adept at AI cost management maintain evaluation benchmarks that reflect their actual workloads. These metrics let teams test new models consistently as they enter the market, regardless of the AI harness in use. He expects more of this work to shift to service providers in the future.

Summary of findings

  • SemiAnalysis claims Anthropic's subscription plans offer roughly five times the API-equivalent value of OpenAI's mid-tier models.
  • Claude subscriptions consume about 42 percent of Anthropic's inference compute while producing only 10 percent of its revenue.
  • Claude Pro ($20/month) delivers 2.9 billion tokens per month, compared to ChatGPT's 1 billion tokens at the same price.
  • Artificial Analysis estimates GPT-6.1 Sol at $0.72 per task and Claude Opus 5.5 at $5.98 per task.
  • Enterprise AI costs are rising. Ninety-three percent of enterprises exceeded their AI budgets in a July McKinsey report.
  • Some teams reduce costs by routing workloads to open source models via model routers like Skydive Glide, with reported savings of 75 percent or more.
  • Maintaining workload-specific evaluation benchmarks helps teams test new models consistently and control costs.

"When you do something, you should burn yourself completely, like a good bonfire, leaving no trace of yourself." — Shunryu Suzuki