Token Pricing Reference — Major LLM Providers
Snapshot of
data/pricing.json(v2026.07) — reviewed 2026-07-26. This file is regenerated from the dataset. To propose a correction, editdata/pricing.jsonand re-run the generator. Live changelog: data/pricing-changelog.md. SLA: Reviewed monthly. Provider pricing changes reflected within 7 days. Every figure carries a verification date and a source link.
Pricing Tables
OpenAI
| Model | Input ($/1M) | Cached Input ($/1M) | Output ($/1M) | Context | Tier | Source |
|---|---|---|---|---|---|---|
| GPT-5 | $1.25 | $0.125 | $10 | 400K | Frontier | verified 2026-07-26 |
| GPT-5 Mini | $0.25 | $0.025 | $2 | 400K | Mid | verified 2026-07-26 |
| GPT-5 Nano | $0.05 | $0.005 | $0.4 | 128K | Cheap | verified 2026-07-26 |
| o3 | $2 | — | $8 | 200K | Reasoning | verified 2026-07-26 |
Caching: Automatic caching on repeated context; ~75-90% off cached input on current-gen models.
Batch API: 50% off standard rates, ~24h SLA.
Anthropic
| Model | Input ($/1M) | Cached Input ($/1M) | Output ($/1M) | Context | Tier | Source |
|---|---|---|---|---|---|---|
| Claude Opus 4.5 | $5 | $0.5 | $25 | 200K | Frontier | verified 2026-07-26 |
| Claude Sonnet 5 | $2 | $0.2 | $10 | 200K | Frontier | verified 2026-07-26 |
| Claude Haiku 4.5 | $1 | $0.1 | $5 | 200K | Mid | verified 2026-07-26 |
Caching: Explicit cache control; 5-min TTL default. Cache writes billed at 1.25x (5-min) or 2.0x (1-hr) base input; cache reads at ~10% of base input.
Batch API: 50% off standard rates, ~24h SLA.
| Model | Input ($/1M) | Cached Input ($/1M) | Output ($/1M) | Context | Tier | Source |
|---|---|---|---|---|---|---|
| Gemini 3 Pro | $2 / $3.5 (>200K ctx) | — | $12 / $14 (>200K ctx) | 1M | Frontier | verified 2026-07-26 |
| Gemini 2.5 Flash | $0.15 / $0.3 (>200K ctx) | — | $0.6 / $1.8 (>200K ctx) | 1M | Mid | verified 2026-07-26 |
| Gemini 2.0 Flash Lite | $0.075 | — | $0.3 | 1M | Cheap | verified 2026-07-26 |
Caching: Implicit automatic caching on Gemini 2.5+; discount surfaces on the invoice.
DeepSeek
| Model | Input ($/1M) | Cached Input ($/1M) | Output ($/1M) | Context | Tier | Source |
|---|---|---|---|---|---|---|
| DeepSeek V3.2 | $0.28 | $0.028 | $0.42 | 128K | Cheap | verified 2026-07-26 |
Caching: Disk-based cache; ~90% discount on cached input.
Meta Llama (hosted)
| Model | Input ($/1M) | Cached Input ($/1M) | Output ($/1M) | Context | Tier | Source |
|---|---|---|---|---|---|---|
| Llama 4 Scout | $0.15 | — | $0.4 | 10M | Cheap | verified 2026-07-26 |
| Llama 4 Maverick | $0.35 | — | $1.4 | 1M | Mid | verified 2026-07-26 |
Caching: Varies by host.
Mistral
| Model | Input ($/1M) | Cached Input ($/1M) | Output ($/1M) | Context | Tier | Source |
|---|---|---|---|---|---|---|
| Mistral Large | $2 | — | $6 | 128K | Frontier | verified 2026-07-26 |
| Mistral Medium | $0.4 | — | $1.2 | 128K | Mid | verified 2026-07-26 |
How to price a request
cost_per_request = (input_tokens × input_$/M ÷ 1_000_000)
+ (output_tokens × output_$/M ÷ 1_000_000)
With prompt caching on the input:
cost_per_request ≈ ((1-hit_rate) × input_tokens × input_$/M
+ hit_rate × input_tokens × cached_input_$/M
+ output_tokens × output_$/M) ÷ 1_000_000
The site's calculators (/calculator, /hub) apply these formulas against this same dataset.