Models and pricing
Console pricing is authoritative
Prices below apply the current production group multipliers and are provided for estimation. Check the Sakrylle Console for model availability, balance, and actual charges.
Group overview
One API key binds to exactly one group. Create separate keys in the Console when you need models from different groups.
| Group | Multiplier | Billing | Notes |
|---|---|---|---|
| Claude-Max | 1.3 | token | 9 Claude models, Claude Code only |
| Claude-Max-C | 1.5 | token | Same as Claude-Max, external API supported |
| GPT-Pro | 0.4 | token | GPT-6 Astra, 5.6, 5.5, and 5.4 families |
| GPT-Pro-Special | 0.28 | token | Same as GPT-Pro, separate key channel |
| GPT-Plus | 0.18 | token | Same model set as GPT-Pro |
| GPT-Team | 0.12 | token | Same model set as GPT-Pro |
| GPT-Image | 1.0 | per request | 4 image models, ￥0.15/call |
| Deepseek-Anthropic | 1.0 | peak/off-peak token | DeepSeek V4, Anthropic-native protocol |
| Deepseek-OpenAI | 1.0 | peak/off-peak token | DeepSeek V4, OpenAI-compatible protocol |
| DeepSeek-Plan-Anthropic | 0.6 | peak/off-peak token | DeepSeek Reverse, Anthropic-native protocol |
| DeepSeek-Plan-OpenAI | 0.6 | peak/off-peak token | DeepSeek Reverse, OpenAI-compatible protocol |
| Grok-API | 0.1 | token / per request | 2 text models and 3 image models |
| Google-Gemini | 0.4 | token | 6 Gemini models, OpenAI-compatible protocol |
| Grok-Build | 0.4 | token | grok-4.5 and grok-4.6 |
Billing rules
Token-billed groups first resolve the channel price in effect at the request start time, then apply the group multiplier. All four DeepSeek groups currently use time-based prices; other token groups use a fixed backend base price.
actual = input token cost + output token cost + cache cost
each cost = tokens / 1,000,000 × resolved backend price × group multiplierThe token tables below already apply each group multiplier and use ￥ / MTok. The image group bills only successful requests; failed requests are not charged.
Claude pricing
Claude-Max and Claude-Max-C offer the same nine Claude models and support a 1M context window.
Claude-Max
Claude-Max is restricted to Claude Code.
| Model | Input | Output | Cache read | Cache write |
|---|---|---|---|---|
claude-opus-4-8 / 4-7 / 4-6 / 5 | ￥6.5 | ￥32.5 | ￥0.65 | ￥8.125 |
claude-sonnet-5 | ￥2.6 | ￥13 | ￥0.26 | ￥3.25 |
claude-sonnet-4-6 | ￥3.9 | ￥19.5 | ￥0.39 | ￥4.875 |
claude-haiku-4-5-20251001 | ￥1.3 | ￥6.5 | ￥0.13 | ￥1.625 |
claude-fable-5 | ￥13 | ￥65 | ￥1.3 | ￥16.25 |
claude-fable-5-1 | ￥13 | ￥65 | ￥1.3 | ￥16.25 |
Claude-Max-C
Claude-Max-C supports external API clients.
| Model | Input | Output | Cache read | Cache write |
|---|---|---|---|---|
claude-opus-4-8 / 4-7 / 4-6 / 5 | ￥7.5 | ￥37.5 | ￥0.75 | ￥9.375 |
claude-sonnet-5 | ￥3 | ￥15 | ￥0.3 | ￥3.75 |
claude-sonnet-4-6 | ￥4.5 | ￥22.5 | ￥0.45 | ￥5.625 |
claude-haiku-4-5-20251001 | ￥1.5 | ￥7.5 | ￥0.15 | ￥1.875 |
claude-fable-5 | ￥15 | ￥75 | ￥1.5 | ￥18.75 |
claude-fable-5-1 | ￥15 | ￥75 | ￥1.5 | ￥18.75 |
GPT pricing
GPT-Pro
| Model | Input | Output | Cache read | Cache write |
|---|---|---|---|---|
gpt-6-astra / gpt-6 | ￥4 | ￥20 | ￥0.4 | ￥5 |
gpt-5.6-sol, codex-auto-review | ￥2 | ￥12 | ￥0.2 | ￥2.5 |
gpt-5.6-terra | ￥0.8 | ￥4.8 | ￥0.08 | ￥1 |
gpt-5.5 | ￥2 | ￥12 | ￥0.2 | Not billed |
gpt-5.4 | ￥1 | ￥6 | ￥0.1 | Not billed |
GPT-Pro-Special
| Model | Input | Output | Cache read | Cache write |
|---|---|---|---|---|
gpt-6-astra / gpt-6 | ￥2.8 | ￥14 | ￥0.28 | ￥3.5 |
gpt-5.6-sol, codex-auto-review | ￥1.4 | ￥8.4 | ￥0.14 | ￥1.75 |
gpt-5.6-terra | ￥0.56 | ￥3.36 | ￥0.056 | ￥0.7 |
gpt-5.5 | ￥1.4 | ￥8.4 | ￥0.14 | Not billed |
gpt-5.4 | ￥0.7 | ￥4.2 | ￥0.07 | Not billed |
GPT-Plus
| Model | Input | Output | Cache read | Cache write |
|---|---|---|---|---|
gpt-6-astra / gpt-6 | ￥1.8 | ￥9 | ￥0.18 | ￥2.25 |
gpt-5.6-sol, codex-auto-review | ￥0.9 | ￥5.4 | ￥0.09 | ￥1.125 |
gpt-5.6-terra | ￥0.36 | ￥2.16 | ￥0.036 | ￥0.45 |
gpt-5.5 | ￥0.9 | ￥5.4 | ￥0.09 | Not billed |
gpt-5.4 | ￥0.45 | ￥2.7 | ￥0.045 | Not billed |
GPT-Team
| Model | Input | Output | Cache read | Cache write |
|---|---|---|---|---|
gpt-6-astra / gpt-6 | ￥1.2 | ￥6 | ￥0.12 | ￥1.5 |
gpt-5.6-sol, codex-auto-review | ￥0.6 | ￥3.6 | ￥0.06 | ￥0.75 |
gpt-5.6-terra | ￥0.24 | ￥1.44 | ￥0.024 | ￥0.3 |
gpt-5.5 | ￥0.6 | ￥3.6 | ￥0.06 | Not billed |
gpt-5.4 | ￥0.3 | ￥1.8 | ￥0.03 | Not billed |
GPT 5.6 models bill cache-write tokens at 125% of the corresponding input price. GPT 5.5 and 5.4 models currently do not bill cache writes. Cache reads cost 10% of the corresponding input price for all GPT models.
Long context billing
All models listed above across the four GPT groups, their compact names, and codex-auto-review use long-context billing. The trigger total is calculated as follows:
gpt-6-astra / gpt-6 use a backend base price of ￥10 input, ￥50 output, ￥1 cache read, and ￥12.5 cache write per MTok. Above 272K, the 2.0x input-side and 1.5x output multipliers above apply.
trigger total = input tokens + cache-write tokens + cache-read tokensWhen the trigger total is strictly greater than 272,000 tokens, the following multipliers apply to the entire request:
| Charge | Long-context multiplier |
|---|---|
| Input | 2.0x |
| Cache write | 2.0x |
| Cache read | 2.0x |
| Output | 1.5x |
The multipliers apply to the whole request, not only the portion above 272,000 tokens. The tables above show standard-context prices with the group multiplier already applied; after the threshold is crossed, multiply each listed price only by the corresponding long-context multiplier.
Compact mappings
Compact names use the same price as their mapped standard model:
| Compact model | Maps to |
|---|---|
gpt-5.6-sol-openai-compact | gpt-5.6-sol |
gpt-5.6-terra-openai-compact | gpt-5.6-terra |
gpt-5.5-openai-compact | gpt-5.5 |
gpt-5.4-openai-compact | gpt-5.4 |
codex-auto-review rewrites to gpt-5.6-sol and uses the gpt-5.6-sol price.
Gemini pricing
Google-Gemini uses OpenAI-compatible endpoints.
| Model | Input | Output | Cache read |
|---|---|---|---|
gemini-3-pro-preview / gemini-3.1-pro / gemini-3.1-pro-preview | ￥0.8 | ￥4.8 | ￥0.08 |
gemini-3.5-flash | ￥0.6 | ￥3.6 | ￥0.06 |
gemini-3.6-flash | ￥0.6 | ￥3 | ￥0.06 |
gemini-3.7-flash | ￥0.3 | ￥1.5 | ￥0.03 |
Image pricing
| Group | Model | Price |
|---|---|---|
| GPT-Image | gpt-image-2 | ￥0.15 / call |
| GPT-Image | gpt-image-2.5 | ￥0.15 / call |
| GPT-Image | gpt-image-2.5-flare | ￥0.15 / call |
| GPT-Image | gpt-image-2.5-sunburst | ￥0.15 / call |
Image generation and editing are billed per successful call. See the Images API for request formats.
DeepSeek pricing
All four DeepSeek groups use peak/off-peak pricing, but official and Plan groups have different peak days. Use Deepseek-Anthropic or DeepSeek-Plan-Anthropic with the Anthropic-native /v1/messages protocol; use Deepseek-OpenAI or DeepSeek-Plan-OpenAI with OpenAI-compatible endpoints.
Effective time
Peak/off-peak pricing takes effect at 2026-08-17 00:00 Asia/Shanghai. Before then, each price row uses its old backend base price; the tables below show the currently effective peak/off-peak prices.
The price tier is locked from the request start time in the Asia/Shanghai timezone. A long-running request does not change price when it crosses a time boundary.
| Groups | Day and time | Tier | Time multiplier |
|---|---|---|---|
| Official | Monday-Friday, 09:00 to 12:00 | Peak | 1.0 |
| Official | Monday-Friday, 14:00 to 18:00 | Peak | 1.0 |
| Official | Other weekday hours and all weekend | Off-peak | 0.5 |
| Plan | Every day, 09:00 to 12:00 | Peak | 1.0 |
| Plan | Every day, 14:00 to 18:00 | Peak | 1.0 |
| Plan | All other daily hours | Off-peak | 0.5 |
Windows are left-closed and right-open: 09:00 and 14:00 are peak, while 12:00 and 18:00 are already off-peak. The following actual prices include both the time multiplier and group multiplier, in ￥ / MTok. These groups currently do not charge for cache writes.
Official groups
Deepseek-Anthropic and Deepseek-OpenAI have the same prices; only the protocol differs. Flash and Flash Vision use the same prices.
| Model | Tier | Input | Output | Cache read |
|---|---|---|---|---|
deepseek-v4-flash / deepseek-v4-flash-vision-exp | Peak | ￥3 | ￥9 | ￥0.1 |
deepseek-v4-flash / deepseek-v4-flash-vision-exp | Off-peak | ￥1.5 | ￥4.5 | ￥0.05 |
deepseek-v4-pro | Peak | ￥9 | ￥27 | ￥0.3 |
deepseek-v4-pro | Off-peak | ￥4.5 | ￥13.5 | ￥0.15 |
Plan groups
DeepSeek-Plan-Anthropic and DeepSeek-Plan-OpenAI use the DeepSeek-Reverse channel and have the same prices; only the protocol differs. Both groups now use the schedule above and offer Flash, Flash-0731, Pro, and Pro-0813:
| Model | Tier | Input | Output | Cache read |
|---|---|---|---|---|
deepseek-v4-flash / deepseek-v4-flash-0731 | Peak | ￥1.8 | ￥5.4 | ￥0.06 |
deepseek-v4-flash / deepseek-v4-flash-0731 | Off-peak | ￥0.9 | ￥2.7 | ￥0.03 |
deepseek-v4-pro / deepseek-v4-pro-0813 | Peak | ￥0.792 | ￥2.376 | ￥0.0264 |
deepseek-v4-pro / deepseek-v4-pro-0813 | Off-peak | ￥0.396 | ￥1.188 | ￥0.0132 |
Grok pricing
| Model | Input | Output | Cache read |
|---|---|---|---|
grok-4.5 | ￥0.2 | ￥0.6 | ￥0.05 |
grok-4.6 | ￥0.2 | ￥0.6 | ￥0.05 |
The grok-imagine-image, grok-imagine-image-2.0, and grok-imagine-image-quality models in Grok-API cost ￥0.02 per successful call (￥0.20 channel base price × the 0.1 group multiplier).
Grok-Build
| Model | Input | Output | Cache read |
|---|---|---|---|
grok-4.5 | ￥0.8 | ￥2.4 | ￥0.2 |
grok-4.6 | ￥0.8 | ￥2.4 | ￥0.2 |
Streaming is strongly recommended for Grok. Clients should retry 429 and 503 responses with exponential backoff.
Volume discounts
For volume use or other billing questions, contact support@sakrylle.com.