LLM API pricing, side by side
Token prices for six providers, taken from their official pricing pages and nothing else. Standard tier, USD per 1 million tokens, with the caching and long-context caveats that quietly change your bill.
Runs entirely in your browser. Nothing you enter leaves this page.
Every number verified against the linked source on 2026-10-16. The spread is real: Claude Haiku 5.5 costs $0.60 per combined 1M in + 1M out, GPT-5.5 Pro costs $210.00.
Anthropic
official pricing page| Model | Input /1M | Cached /1M | Output /1M | 1M in + 1M out | Context |
|---|---|---|---|---|---|
| Claude Fable 5 | $10.00 | $1.00 | $50.00 | $60.00 | 1M |
| Claude Fable 5.1 | $10.00 | $0.25 | $50.00 | $60.00 | 1M |
| Claude Mythos 5 | $10.00 | $1.00 | $50.00 | $60.00 | 1M |
| Claude Mythos 5.1 | $10.00 | $0.25 | $50.00 | $60.00 | 1M |
| Claude Opus 4.8 | $5.00 | $0.50 | $25.00 | $30.00 | 1M |
| Claude Opus 5 | $5.00 | $0.50 | $25.00 | $30.00 | 1M |
| Claude Opus 5.5 | $4.00 | $0.20 | $20.00 | $24.00 | 1M |
| Claude Sonnet 4.6 | $3.00 | $0.30 | $15.00 | $18.00 | 1M |
| Claude Sonnet 5 | $2.00 | $0.20 | $10.00 | $12.00 | 1M |
| Claude Sonnet 5.5 | $2.00 | $0.10 | $10.00 | $12.00 | 1M |
| Claude Haiku 4.5 | $1.00 | $0.10 | $5.00 | $6.00 | - |
| Claude Haiku 5.5tiered | $0.10 | $0.01 | $0.50 | $0.60 | 1M |
- Cached input is the cache read price, "0.1x base input price (0.025x on Claude Fable 5.1 and Claude Mythos 5.1; 0.05x on Claude Opus 5.5 and Claude Sonnet 5.5)". A cache hit is $0.25 per million tokens on those first two, $0.20 on Claude Opus 5.5 and $0.10 on Claude Sonnet 5.5. Cache writes cost extra: 1.25x input for 5-minute TTL, 2x for 1-hour TTL.
- Batch API: 50 percent off input and output.
- Claude Sonnet 5's $2 in / $10 out was announced as introductory pricing through 2026-08-31. Anthropic has since made it the standard price and cancelled the increase to $3 / $15 that was scheduled for 2026-09-01.
- Claude 4.6 and later models (except Claude Haiku 5.5), plus Claude Mythos Preview, include the full 1M-token context window at standard pricing. Anthropic states the rule by generation rather than by model list.
- Claude Haiku 5.5 is priced by prompt length. Prompts up to 100,000 tokens pay the rates shown; a prompt over 100,000 tokens pays $0.50 in / $2.50 out per million tokens, with cache reads at $0.05.
- Pinning inference to the US with the inference_geo parameter costs 1.1x on every token category, including cache writes and reads, on Claude 4.6 and later. Global routing is the default and is priced as shown.
- Claude 4.7 and later models, which includes Opus 5.5, Opus 5, Sonnet 5 and the Fable and Mythos 5 families, use a newer tokenizer that Anthropic says produces roughly 30% more tokens for the same text. Sonnet 4.6 and earlier use the previous one, so a lower price per million tokens does not guarantee a lower bill for identical work.
OpenAI
official pricing page| Model | Input /1M | Cached /1M | Output /1M | 1M in + 1M out | Context |
|---|---|---|---|---|---|
| GPT-5.5 Protiered | $30.00 | - | $180.00 | $210.00 | - |
| GPT 5.6 Cyber | $12.50 | $1.25 | $75.00 | $87.50 | - |
| GPT-6 Astratiered | $10.00 | $1.00 | $50.00 | $60.00 | - |
| GPT-5.5tiered | $5.00 | $0.50 | $30.00 | $35.00 | - |
| GPT Rosalind Discovery | $5.00 | $0.50 | $25.00 | $30.00 | - |
| GPT Rosalind Research | $5.00 | $0.50 | $25.00 | $30.00 | - |
| GPT-5.6 Soltiered | $4.00 | $0.40 | $20.00 | $24.00 | - |
| GPT-5.4tiered | $2.50 | $0.25 | $15.00 | $17.50 | - |
| GPT-5.6 Terratiered | $2.00 | $0.20 | $12.00 | $14.00 | - |
| GPT-6 Soltiered | $2.00 | $0.20 | $10.00 | $12.00 | - |
| GPT 6.1 Soltiered | $2.00 | $0.10 | $10.00 | $12.00 | - |
| GPT-5.3-Codex | $1.75 | $0.175 | $14.00 | $15.75 | - |
| GPT-5.4 mini | $0.75 | $0.075 | $4.50 | $5.25 | - |
| GPT-5.4 nano | $0.20 | $0.02 | $1.25 | $1.45 | - |
| GPT-5.6 Lunatiered | $0.20 | $0.02 | $1.20 | $1.40 | - |
| GPT-6 Lunatiered | $0.10 | $0.01 | $0.50 | $0.60 | - |
- The GPT-6 models and the flagship GPT-5.6 models each carry a separate long-context band on the pricing page, at 2x input, 2x cached input and 1.5x output. GPT-6 Astra's long-context band is $20.00 in / $2.00 cached / $75.00 out. A footnote under the flagship tables defines the bands: "Short context: ≤272K input tokens. Long context: >272K input tokens." The separate Cyber models table, which also lists gpt-5.6-sol, uses the same short and long context columns. Prior-generation rows now sit behind the page's "All models" view rather than on the flagship table.
- Batch and Flex: 50 percent off. Fast mode, renamed from Priority processing on 2026-07-30, costs 2x standard.
- Regional processing (data residency) endpoints carry a 10% uplift for models released on or after 2026-03-05, and FedRAMP endpoints carry the same 10% uplift. GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna support EU data residency with Standard, Flex and Batch processing, and in Fast mode; Fast mode is not available with EU data residency for GPT-6 Astra.
| Model | Input /1M | Cached /1M | Output /1M | 1M in + 1M out | Context |
|---|---|---|---|---|---|
| Gemini 3.1 Pro (preview)tiered | $2.00 | $0.20 | $12.00 | $14.00 | - |
| Gemini 4 Argon | $2.00 | $0.10 | $10.00 | $12.00 | - |
| Gemini 2.5 Protiered | $1.25 | $0.125 | $10.00 | $11.25 | - |
| Gemini 3.6 Flash | $0.75 | $0.075 | $3.75 | $4.50 | - |
| Gemini 3.8 Flash | $0.75 | $0.075 | $3.75 | $4.50 | - |
| Gemini 3.5 Flash-Lite | $0.30 | $0.03 | $2.50 | $2.80 | - |
- Gemini 3.1 Pro and 2.5 Pro charge higher rates for prompts above 200K tokens (3.1 Pro: $4.00 in / $18.00 out).
- Gemini 3.5 Flash-Lite's input price covers every modality: its cell reads "$0.30 (text / image / video / audio)". The other Flash rows shown list one unqualified input price. The separate, higher audio input rates on Google's page belong to models this table does not carry, such as Gemini 2.5 Flash and the Live and TTS models.
- Context caching storage is billed per hour on top of cached-input rates. The Pro rows show "$4.50 / 1,000,000 tokens per hour (storage price)"; the Flash rows show $1.00, halved to $0.50 through 2026-12-31 on the promotionally-priced Flash models.
- Gemini 3.6 Flash and 3.8 Flash are promotionally priced at $0.75 in / $3.75 out through 2026-12-31 and rise to $1.50 / $7.50 on 2027-01-01. Google deprecated Gemini 3.5 Flash and 3.7 Flash on 2026-10-08; requests to those ids now route to 3.6 Flash and 3.8 Flash respectively.
- Gemini 4 Argon, GA 2026-10-13, is priced at $2.00 in / $10.00 out / $0.10 cached through 2026-11-30, and every rate doubles on 2026-12-01 ($4.00 / $20.00 / $0.20). It is not available on the Free tier and supports implicit caching only.
DeepSeek
official pricing page| Model | Input /1M | Cached /1M | Output /1M | 1M in + 1M out | Context |
|---|---|---|---|---|---|
| DeepSeek V4 Pro | $0.66 | $0.022 | $1.98 | $2.64 | 1M |
| DeepSeek Flash | $0.15 | $0.003 | $0.60 | $0.75 | 1M |
- Cached input is the cache-hit price; cache misses pay the normal input rate.
- Peak/off-peak pricing took effect 2026-08-16. Rates shown are OFF-PEAK; peak hours (01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays) cost exactly 2x.
- DeepSeek renamed its flash model to deepseek-flash on 2026-09-10, when it began serving DeepSeek-V4.1-Flash, and cut the price to $0.15 in / $0.60 out from $0.22 / $0.66. The retired names deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted; their requests are served by the DeepSeek-V4.1-Flash-1017 snapshot and billed at the flash rate.
- DeepSeek announced on 2026-09-10 that from 2026-09-14 it would route every deepseek-v4-pro request to V4.1 Flash and bill it at the flash price, then withdrew that on the day: on 2026-09-14 its pricing page said it would "continue providing API services for DeepSeek V4 Pro after September 14, 2026, with the billing method remaining unchanged". That notice has since come off the page, which still lists deepseek-v4-pro at the unchanged rates shown.
- The deepseek-chat and deepseek-reasoner aliases were deprecated on 2026-07-24 and no longer appear on DeepSeek's pricing page.
Mistral
official pricing page| Model | Input /1M | Cached /1M | Output /1M | 1M in + 1M out | Context |
|---|---|---|---|---|---|
| Mistral Medium 3.5 | $1.50 | $0.15 | $7.50 | $9.00 | - |
| Mistral Large 4 | $0.68 | $0.07 | $2.09 | $2.77 | 1M |
| Mistral Large 3 | $0.50 | $0.05 | $1.50 | $2.00 | - |
| Codestral | $0.30 | $0.03 | $0.90 | $1.20 | - |
| Mistral Small 4 | $0.15 | $0.015 | $0.60 | $0.75 | - |
- Mistral Large 3 is priced below the newer Mistral Medium 3.5 on the official page ($0.50 / $1.50 against $1.50 / $7.50). Verified twice on 2026-06-10 and again on 2026-08-19, because the ordering reads like an error and is not one.
- Cached input is 0.1x input on every model. Since 2026-10 Mistral publishes it as its own Cached input column on the pricing table.
- Mistral Large 4 launched 2026-10-06 as a Public Preview and is shown at its sale price. Mistral lists $1.36 in / $0.14 cached / $4.18 out struck through and $0.68 / $0.07 / $2.09 as the sale price, and states no end date for the sale.
- Batch API: 50 percent off, and regional inference adds 10 percent.
| Model | Input /1M | Cached /1M | Output /1M | 1M in + 1M out | Context |
|---|---|---|---|---|---|
| Grok 4.5tiered | $2.00 | $0.30 | $6.00 | $8.00 | 500K |
| Grok 4.6tiered | $2.00 | $0.50 | $6.00 | $8.00 | 500K |
| Grok 4.7tiered | $2.00 | $0.50 | $6.00 | $8.00 | 500K |
| Grok 4.20 (reasoning)tiered | $1.25 | $0.20 | $2.50 | $3.75 | 1M |
| Grok 4.3tiered | $1.25 | $0.20 | $2.50 | $3.75 | 1M |
| Grok Build 0.1tiered | $1.00 | $0.20 | $2.00 | $3.00 | 256K |
- Batch API: 20 percent off, on Grok 4.3 and the Grok 4.20 models only. xAI lists those four model ids at 20 percent and then states "Models not listed above have no batch discount", so Grok 4.7, Grok 4.6, Grok 4.5 and Grok Build 0.1 get no batch discount.
- Every Grok model shown has a long-context tier at the same threshold, 200K tokens ("Long context >= 200k tokens"). xAI bills "the long context rates for all tokens in a request once its prompt reaches the model's long context threshold", so at or above that threshold input, cached input and output each cost exactly 2x the rate shown.
- Requests sent to xAI's US regional endpoint are billed at 1.1x the global token rates shown, a 10% premium. The endpoint currently serves grok-4.7 and grok-4.6 only.
How to read this table honestly
- Cached input is not one thing. Anthropic and DeepSeek list a cache-read price; Anthropic also bills cache writes (1.25x input), Google bills cache storage per hour. Two providers with the same "cached" number can produce different bills.
- "Tiered" means the listed price is the floor. OpenAI's GPT-5.5/5.4 and Google's Pro models charge roughly double for long-context requests, which is exactly what agent workflows produce.
- Batch discounts are large. Anthropic, OpenAI, and Mistral all offer 50 percent off for async batch workloads; xAI 20 to 50 percent.
- Excluded providers are excluded for a reason. Meta Llama API, Cohere, and Amazon Nova render their prices client-side, so they could not be verified from source. No number on this page is quoted from a third-party blog.
