Breaking

Anthropic, OpenAI Shift Enterprise Customers to Usage-Based Pricing as Compute Costs Climb

June 18, 2026 at 11:37 EDT

Sales & Business
Infra & Chips
Software Dev & Coding

Major AI companies including OpenAI, Anthropic and GitHub are moving their enterprise pricing away from flat per-seat subscriptions toward usage-based billing. The shift, driven by surging inference compute costs, has accelerated from late 2025 into the spring of 2026.

April 2026 · Enterprise AI Pricing

The End of the Flat Fee: AI Vendors Switch to Pay-Per-Token

OpenAI, Anthropic and others are unbundling fixed per-seat subscriptions, charging a small access fee plus separate, metered token consumption — a response to soaring inference costs that flat plans can no longer absorb.

Anthropic Claude Enterprise: the model flips

Before

$200/user/mo

Fixed seat fee with tokens bundled in — heavy users effectively subsidized.

→

After

~$20/user/mo

Access fee only + every token billed at API rates on top.

of heavy users consume half of all compute

~$6/day

average Claude Code developer spend

90%

of developers stay under $12/day

What a token costs now

Output price per million tokens — the metered charge that now scales with usage.

Haiku 4.5

$15

Sonnet 4.6

$25

Opus 4.6

Block height ∝ price per million output tokens. Opus output costs 5× Haiku.

Why flat fees broke

Coding agents

Claude Code, Codex

→

Long sessions

large context windows

→

Token surge

inference costs explode

→

Meter it

pay-per-use billing

Seen as more honest

A rational acknowledgment that VC-subsidized flat rates were unsustainable. You pay for what you actually compute.

Sticker shock for heavy users

Claude Code–reliant teams face unexpected bills. Mitigations: clear context, prompt caching, off-peak use, evaluate rivals.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.