Estimate tokens, compare official API prices, model multi-turn conversations and project costs from one request to annual scale. Every displayed rate links to its first-party source.
Model-family estimate. Tokenization varies by model, language, JSON, emoji and code. Use Known Tokens when an exact provider count is available.
Cheapest estimate
$0.00
Selected models
4
Input tokens
0
Output tokens
0
| GPT-5.6 Sol | $0.00 |
|---|---|
| GPT-5.6 Terra | $0.00 |
| GPT-5.6 Luna | $0.00 |
| Claude Fable 5 | $0.00 |
USD per million tokens unless noted. Open a first-party source before making a purchasing decision.
| Provider / model | Status | Input | Cached | Output | Context | Request | Monthly | Difference | Source | Verified |
|---|---|---|---|---|---|---|---|---|---|---|
| GPT-5.6 SolOpenAI | current | $5.00 | $0.5000 | $30.00 | 1,050,000 | $0.00 | $0.00 | — | Official | 2026-07-16 |
| GPT-5.6 TerraOpenAI | current | $2.50 | — | $15.00 | 1,050,000 | $0.00 | $0.00 | — | Official | 2026-07-16 |
| GPT-5.6 LunaOpenAI | current | $1.00 | — | $6.00 | 1,050,000 | $0.00 | $0.00 | — | Official | 2026-07-16 |
| Claude Fable 5Anthropic | current | $10.00 | $1.00 | $50.00 | 1,000,000 | $0.00 | $0.00 | — | Official | 2026-07-16 |
| GPT-5.6 Sol | $0.00 |
|---|---|
| GPT-5.6 Terra | $0.00 |
| GPT-5.6 Luna | $0.00 |
| Claude Fable 5 | $0.00 |
GPT-5.6 Sol
| Current | $0.00 |
|---|---|
| 2× | $0.00 |
| 5× | $0.00 |
| 10× | $0.00 |
Rules filter recorded technical eligibility, then show price positions. They do not invent quality scores or call one model objectively best.
Cheapest technically eligible
$0.00/month
Middle-cost eligible option
$0.00/month
Premium-cost eligible option
$0.00/month
Quality, latency, rate limits, regional availability, and workload-specific benchmark performance are not ranked without separately sourced evidence.
Privacy: Your pasted text is processed in your browser and is not stored. Analytics events contain only action names—not prompt text, documents, API keys, or calculation contents.
Providers normally publish rates per million tokens. A request can combine standard input, cached reads, cache writes, model output and separately billed tools. Long-context thresholds may change the price for every token in the request. For that reason, a single blended “price per token” can be misleading.
standard input tokens / 1,000,000 × input rate + cached tokens / 1,000,000 × cache rate + output tokens / 1,000,000 × output rate + tool and modality charges = total
PiSkill uses unrounded decimal arithmetic internally and rounds only for presentation. Provider billing systems and taxes may differ.
Official OpenAI model pages; prices are API usage, not ChatGPT subscriptions.
Official Claude Platform pricing, including cache-write and dated promotional rates.
Official Gemini Developer API paid-tier rates.
Official xAI pricing, including long-context and server-side tool fees.
Official Mistral API pricing; batch is 50% and eligible cached input is 90% lower.
Official DeepSeek V4 API pricing and documented context-cache rates.
Token prices depend on the provider, model, direction and pricing tier. Input, cached input and output can have different rates. This calculator applies each selected model's sourced USD rates to the same workload.
Multiply each billable token category by its applicable per-million rate, then add tool and modality charges. Monthly cost multiplies the per-request result by request volume; the full formula remains visible beside each result.
There is no universal cheapest model. The answer changes with output length, cache use, batch eligibility, context size, tools and technical requirements. PiSkill identifies the lowest calculated price only among eligible selected models.
Cost arithmetic can be exact when token counts and pricing conditions are known. Pasted-text counts are estimates unless a maintained local tokenizer matches the selected model. Provider invoices remain authoritative.
Input tokens contain the instructions and context sent to a model. Output tokens are generated by the model. Providers commonly price output tokens at a higher rate.
Prompt caching can reuse stable input across requests. Supported providers may charge separate cache-read and cache-write rates, with eligibility, retention and minimum-length rules.
A consumer chat subscription and API billing are separate products. Do not assume a chat plan includes an API allowance unless the provider explicitly documents one.
It depends on workload and product needs. API usage is metered; a chat subscription provides a hosted interface with documented usage policies. The comparison is a workload estimate, not a conversion of messages into API tokens.
When an application resends full history, later turns contain more input tokens. PiSkill calculates every turn independently so the cumulative growth is visible.
Measure real input and output, choose the smallest technically suitable model, cache stable prompts, batch delay-tolerant work, trim history, summarize context and verify tool-call charges.