Cycles

Appearance

Claude vs GPT Cost Calculator ​

A free interactive calculator that compares per-call, per-day, per-month, and per-year cost across the major Claude and OpenAI models for any token volume and call rate.

Tip: Open fullscreen ↗ for a wider table, share/export buttons, and a shareable URL that preserves your configuration. The same toolbar is also available below.

ModelInput $ / MOutput $ / MPer callPer dayPer monthPer year
$0.0340$340$10,200$124,100
$0.0170$170$5,100$62,050
$0.0068$68.00$2,040$24,820
$0.0014$14.00$420$5,110
$0.0600$600$18,000$219,000
$0.0300$300$9,000$109,500
$0.0120$120$3,600$43,800
$0.0060$60.00$1,800$21,900

Defaults reflect published list pricing as of 2026-07 from OpenAI and Anthropic; verify before relying on these for budgeting and edit any cell to match your contracted rate. Calculation: (input_tokens × input_$/M + output_tokens × output_$/M) ÷ 1,000,000 × calls/day. Excludes prompt caching, batch discounts, fine-tuning, fast-mode, and data-residency premiums. Claude Sonnet 5 is at introductory pricing ($2 / $10 per M) through 2026-08-31 ($3 / $15 from 2026-09-01). Anthropic's newest models (Fable 5, Opus 4.8, Sonnet 5) use a newer tokenizer that produces ~30% more tokens for the same text than earlier Claude models, so equal token counts across different tokenizers here are not an equal-text comparison.

How the calculation works ​

Model prices change frequently. The defaults are starting points, not a pricing guarantee — edit any rate to match the current provider pricing page or your contracted rates.

The cost for a single LLM call is:

cost_per_call = (input_tokens × input_price_per_M + output_tokens × output_price_per_M) ÷ 1,000,000

Per-day, per-month, and per-year columns multiply by calls_per_day, then by 30 and 365 respectively.

The pricing rates default to widely-published reference values, but they are user-editable above so you can plug in current or contracted rates. The cheapest per-year row is highlighted.

What the calculator does not include ​

  • Prompt caching discounts. Anthropic's cache_control blocks bill cache hits at a fraction of base input pricing (for example, Opus 4.8 cache hits at $0.50 / MTok vs $5 / MTok base input). OpenAI also offers cached-input and Batch API pricing on supported models, but the discount varies by model. If a meaningful share of your prompts are reused, your real cost may be 30–60% lower than the calculator suggests.
  • Batch API discounts. Both providers offer batch-processing discounts (Anthropic at 50%; OpenAI varies).
  • Fine-tuning costs. Per-token rates differ for fine-tuned model variants.
  • Reserved or committed-use pricing. Enterprise contracts often beat list pricing materially.
  • Context-window pricing tiers. Some providers charge differently for very long contexts.

For accurate enterprise planning, treat the calculator as a directional estimate and verify with your provider account manager.

Why estimates are not the same as runtime authority ​

A common pattern: a team uses a calculator like this to project monthly cost at $4,000, sets up an alert for "$5,000 exceeded," and then loses $40,000 in a weekend to an agent that loops while the on-call team sleeps.

Calculators answer "what *might this workload cost?" Cycles can enforce caller-configured budgets at a mandatory application boundary before protected work begins. Application authorization separately decides which calls and tool arguments are allowed. Cost is one dimension of the resulting control design:

  • Blast radius. A single agent action — a deploy, an email blast, a database mutation — can cost more in damage than the agent's entire month of LLM bills. The host must authorize the tool and arguments; Cycles can meter caller-assigned action exposure alongside that decision.
  • Risk-weighted budgeting. Not every tool call is equal. RISK_POINTS lets the caller meter a team-defined exposure score separately from token cost; it is not an authorization decision.
  • Multi-tenant boundaries. A noisy tenant cannot drain shared headroom from quiet ones. See Multi-tenant SaaS.

If your projected $4,000/month is the actual constraint, the right response is not an alert at $5,000 but a pre-execution gate that will not let calls proceed beyond the cap — and an action-authority layer that prevents the catastrophic single mistake that no cost calculator can predict.