The SI Price Index
Public list prices, per 1 million tokens, for the major hosted model APIs — read from each provider's own pricing page and dated below. One row is measured directly on our own hardware rather than quoted from a vendor.
| Provider ▼ | Model | Input $ / 1M tokens | Output $ / 1M tokens | Checked |
|---|---|---|---|---|
| xAI |
Grok 4.7
Global/standard endpoint rate, prompts under 200k tokens (cached input $0.50/1M). Above 200k: $4.00 in / $12.00 out per 1M. The US regional endpoint is 1.1x this rate ($2.20/$6.60 below 200k).
|
$2 | $6 | 2026-10-05 |
| OpenAI |
GPT-6 Astra
Flagship tier, standard pricing.
|
$10 | $50 | 2026-10-05 |
| OpenAI |
GPT-6.1 Sol
Mid tier, standard pricing.
|
$2 | $10 | 2026-10-05 |
| OpenAI |
GPT-6 Luna
Smallest/cheapest current tier, standard pricing.
|
$0.10 | $0.50 | 2026-10-05 |
| OpenAI |
GPT-5
Prior-generation reference point, standard pricing.
|
$1.25 | $10 | 2026-10-05 |
| Mistral |
Mistral Large
Standard pricing. Batch processing is 50% off; cached input tokens up to 90% off.
|
$0.50 | $1.5 | 2026-10-05 |
|
Gemini 3.1 Pro Preview
Prompts under 200k tokens. Above 200k: $4.00 in / $18.00 out per 1M.
|
$2 | $12 | 2026-10-05 | |
|
Gemini 3.8 Flash
Introductory pricing through 31 Dec 2026; rises to $1.50 in / $7.50 out from 1 Jan 2027.
|
$0.75 | $3.75 | 2026-10-05 | |
| DeepSeek |
DeepSeek-V4.1-Flash
Peak-hour, cache-miss rate. Off-peak (nights/weekends UTC) is half price; cache-hit input is far cheaper.
|
$0.30 | $1.2 | 2026-10-05 |
| DeepSeek |
DeepSeek-V4-Pro-0813
Peak-hour, cache-miss rate. Off-peak (nights/weekends UTC) is half price; cache-hit input is far cheaper.
|
$1.32 | $3.96 | 2026-10-05 |
| Anthropic |
Fable 5.1
Flagship tier, built for long-running agents. Standard pricing; US-only inference is 1.1x.
|
$10 | $50 | 2026-10-05 |
| Anthropic |
Opus 5.5
Daily-driver tier for agentic coding and enterprise work. Standard pricing.
|
$4 | $20 | 2026-10-05 |
| Anthropic |
Sonnet 5.5
Mid tier for coding and agents. Standard pricing.
|
$2 | $10 | 2026-10-05 |
| Anthropic |
Haiku 4.5
Fastest, most cost-efficient current tier. Standard pricing.
|
$1 | $5 | 2026-10-05 |
| Our own hardware | Local model | Not measured yet | — | |