LLM Pricing Calculator
Enter token counts, or paste a prompt and let this tool estimate them, and see a real side-by-side dollar comparison across GPT-5.6, Claude Opus 4.7/4.8, and Gemini 3.6 Flash — including each provider's own prompt-caching discount and premium math, which is structured differently across all three.
Last verified
Pricing data checked:
GPT: 0 tokensExact
Claude: 0 tokensEstimated
Gemini: 0 tokensEstimated
- Input
- $0.00
- Cached
- $0.00
- Output
- $0.00
$0.00
- Input
- $0.00
- Cached
- $0.00
- Output
- $0.00
$0.00
Google confirms implicit caching is enabled by default (4,096-token minimum, cost savings passed on automatically) but does not publish an exact discount rate — standard pricing is used below until one is confirmed.
- Input
- $0.00
- Output
- $0.00
$0.00now, through 2026-12-31
$0.00 from 2027-01-01
How this works
Enter your input and output token counts directly, or switch to paste mode and let this tool estimate them from real prompt text — the GPT estimate uses the same tiktoken client-side tokenizer as the AI Token Counter, and Claude/Gemini use the same character-based estimates. Since a response doesn't exist yet at prompt-composition time, output length in paste mode is always a manual number, not something this tool can count for you.
Each provider's caching controls match how that provider actually built caching, not a forced-common shape: OpenAI's single tier applies one write premium and one read discount; Anthropic's two tiers (5- minute, 1-hour) price writes differently depending which one you pick; Gemini's card shows a note instead of a toggle, because Google confirms caching exists but hasn't published an exact discount rate to calculate against.
Pricing reference
| Model | Provider | Input / 1M tokens | Output / 1M tokens | Caching |
|---|---|---|---|---|
| GPT-5.6 Sol (flagship) | OpenAI | $5.00 | $30.00 | 1 tier |
| GPT-5.6 Terra | OpenAI | $2.00 | $12.00 | 1 tier |
| GPT-5.6 Luna | OpenAI | $0.20 | $1.20 | 1 tier |
| GPT-5.6 Cyber | OpenAI | $12.50 | $75.00 | 1 tier |
| Claude Opus 4.7 (standard) | Anthropic | $5.00 | $25.00 | 2 tiers |
| Claude Opus 4.7 (Fast Mode) | Anthropic | $30.00 | $150.00 | 2 tiers |
| Claude Opus 4.8 (standard) | Anthropic | $5.00 | $25.00 | 2 tiers |
| Claude Opus 4.8 (Fast Mode) | Anthropic | $10.00 | $50.00 | 2 tiers |
| Gemini 3.6 Flash (now through 2026-12-31) | $0.75 | $3.75 | Confirmed, rate unpublished | |
| Gemini 3.6 Flash (from 2027-01-01) | $1.50 | $7.50 | Confirmed, rate unpublished |
Sourced directly from each provider's own pricing page as of the "Pricing data checked" date above — see the individual posts in the 2026 LLM Token & Pricing Reset series for the full research and sourcing behind these figures.
Frequently asked
Is this pulling live prices from OpenAI, Anthropic, or Google?
No. None of the three providers publish a public, unauthenticated pricing API, and this tool runs entirely in your browser with no backend — so the figures are a bundled snapshot, checked against each provider's own pricing page as of the date shown above, not a live feed. Re-check against the source links below before a real budgeting decision.
Why does Gemini show two different prices?
Gemini 3.6 Flash's price is scheduled to change on 2027-01-01 (confirmed directly on Google's own pricing page). Showing both the current and future figure side by side, rather than picking one, is deliberate — this exact site previously published the future price as if it were current in an earlier post, and got it wrong. Both numbers are shown so that mistake can't repeat here.
Why is Claude's caching structured differently from GPT-5.6's?
Because the providers actually built it differently, not because this tool is simplifying anything. OpenAI uses one flat cache-write premium and read discount. Anthropic offers two distinct cache tiers (5-minute and 1-hour) with different write premiums. Google confirms caching exists on Gemini but does not publish an exact discount rate anywhere in its documentation, so this tool uses standard pricing for Gemini until one is confirmed rather than guessing at a number.
What does the Claude tokenizer adjustment do?
It only applies in paste-text mode. Claude's tokenizer (since Opus 4.7) counts roughly 30-35% more tokens than older models for the same text, but this tool's Claude estimate is a generic character-based approximation, not Claude's real tokenizer. The adjustment toggle multiplies that estimate by an editable factor (1.325 by default, the midpoint of the reported range) to correct for the gap. It's not applied in manual-number mode, since that mode assumes you already typed in a real, Claude-accurate count.
Does my prompt text get sent anywhere?
No. All calculation happens locally in your browser, the same as the AI Token Counter. Nothing you type or paste is uploaded, logged, or stored.
Emitted as FAQPage JSON-LD from the same frontmatter — one source, no duplicated prose.