TokenCalculator does not scrape a live price API. Each model row carries a checked date. The newest date in the catalog right now is 12 Aug 2026.
How to read the stamp
Rates checked means a human reviewed the official page (or a reviewed sync) on that date. It is a freshness signal, not a same hour guarantee.
What we verify
- Input and output USD per 1 million tokens from the official provider page.
- Cached input, cache write, and batch rates when the provider publishes them.
- Context window and long context tiers when those change the rate.
- A source URL on every model so you can open the same page we used.
What we do not claim
- We do not stream prices in real time.
- We do not invent benchmark scores or community votes.
- We do not model private enterprise amendments or free credit packs.
- Approx token counts are labeled Approx. They are not billed invoices.
Recent catalog notes
| Date | Note |
|---|---|
| 12 Aug 2026 | Checked official OpenAI, Google, and DeepSeek pages plus a reviewed LiteLLM fallback. Confirmed 59 published rates. Added GPT-5.6 Sol, Terra, and Luna. Set Gemini 2.0 Flash Lite cached input and DeepSeek V4 Pro cache hit to the official figures. |
| 2 Aug 2026 | Compare hub expansion: +40 high-intent pairs (frontier/mid, budget, reasoning, host-vs-host Llama/DeepSeek/Qwen/Mixtral, coding, generation migrations). |
| 2 Aug 2026 | Catalog to ~120 models: dated GPT-4o/4.1 and o1-mini/preview, Claude 3.5 Sonnet / 2.1, Gemini 2.0 Pro Exp / Gemma 3, DeepSeek R1 family, Grok 2 Vision, Mixtral 8x22B, Groq SpecDec/QwQ, Command R snapshot, Llama 4 Maverick and R1 on Together/Fireworks. |
| 2 Aug 2026 | Catalog to 100 models: legacy GPT-4 / Claude 3 Haiku / Gemini 1.5 Flash-8B, DeepSeek VL, Grok Vision Beta, Open Mixtral, Gemma 7B, Command R+ snapshot, Llama 3.2 3B on Together/Fireworks. More host and generation compares. |
| 2 Aug 2026 | Week 2: expanded all remaining thin guides with FAQs; +18 models (legacy GPT/Claude/Gemini, DeepSeek Coder, more Groq/Together/Fireworks/Cohere/Mistral). Compare pairs extended for host and generation migrations. |
| 2 Aug 2026 | Week expansion: +20 models (legacy GPT-4 Turbo/o1, Gemini 1.5, DeepSeek V3, Grok 2, Mistral Nemo/Pixtral, more Groq/Together/Fireworks/Cohere, Claude 3.5 Haiku). Bridge guides and pricing guide expanded with FAQs. |
Official sources we check
- OpenAI pricing
- Anthropic pricing
- Google Gemini pricing
- DeepSeek pricing
- xAI pricing
- Mistral pricing
- Groq pricing
- Together pricing
- Fireworks pricing
- Cohere pricing
Before you sign a budget
Open the model page in TokenCalculator, note the checked date, then open the official pricing link. If the two disagree, trust the provider. If you need Claude or Gemini counts for a contract, use that lab’s official count API on the same payload.
Frequently asked questions
Why is the date not today?
We update when we review official pages. A quiet week keeps the last checked date. The stamp is honest on purpose.
How often do you sync?
A weekly script diffs official pages and a reviewed fallback. Changes land after a person checks the diff. That is slower than a scrape and less theatrical.