Amanai Docs

Billing & pricing

One place for billing behavior, current model rates, settlement, and credit packs.

Overview

Amanai uses prepaid credits across chat, media, realtime, embeddings, and rerank. Each endpoint bills in a unit that matches the work delivered; specialized work is not converted into fake chat tokens.

System One structured evaluation costs 0 credits while the offer is active. Currently free for a limited time while the upstream free offering remains available. Free access is not guaranteed and future requests may be priced differently. Usage is still measured.

Current quote wins. Rates can change for future usage as capacity changes. Already settled requests are never recalculated. Use the dashboard calculator or GET /calculate for the quote that applies to a request.

1,000,000,000 credits = Rp150,000 = US$10.

Billing units

API familyBilling unitSettlement basis
ChatInput, cached input, and output tokensUsage reported by the completed or streamed response
EmbeddingsBGE, Qwen3, and Voyage: tokens; EmbeddingGemma: UTF-8 unitsAccepted input with a model-specific minimum
RerankQuery + document tokensAll submitted documents, not only top_n
System OneFree request0 credits; token usage reported
ImagesDelivered imageUsable output images
VideoRequested secondCompleted durable job
Speech synthesisSubmitted charactersPlayable audio delivery
TranscriptionVerified audio durationDelivered transcript
RealtimeSuccessfully established sessionSuccessful WebRTC negotiation

Chat model pricing

Chat credits are rounded up to a whole credit:

bash
credits = ceil(
  (input_tokens - cache_read_tokens) × m_in
  + cache_read_tokens × m_cache
  + output_tokens × m_out
)

m_cache is the rate for input actually reported as cache-read. Cache creation remains input; only cache reads receive the cache multiplier.

ModelInputOutputCache read
GLM 5.3 amanai/glm-5.3 5.6 28 1.4
GLM 5.3 Flash amanai/glm-5.3-flash 2.8 14 0.7
GLM 5.2 amanai/glm-5.2 5.6 28 1.4
GLM 5.1 amanai/glm-5.1 5.6 28 1.4
GLM 5.0 Turbo amanai/glm-5.0-turbo 4.8 24 1.2
GLM 5V Turbo amanai/glm-5v-turbo 4.8 24 1.2
Muse Spark 1.1 amanai/muse-spark-1.1 5 25 1.25
Muse Spark 1.2 amanai/muse-spark-1.2 5 25 1.25
Qwen 3.8 Max amanai/qwen3.8-max 6 30 1.5
Qwen 3.7 Max amanai/qwen3.7-max 4.425 22.125 1.10625
Qwen 3.7 Plus amanai/qwen3.7-plus 2.24 11.2 0.56
Kimi K3 amanai/kimi-k3 9 45 2.25
Kimi K2.7 amanai/kimi-k2.7 3.8 19 0.95
Kimi K2.6 amanai/kimi-k2.6 3.8 19 0.95
Kimi K2.5 amanai/kimi-k2.5 2.28 11.4 0.57
DeepSeek V4 Pro amanai/deepseek-v4-pro 5 25 1.25
DeepSeek V4 Pro 0813 amanai/deepseek-v4-pro-0813 5 25 1.25
DeepSeek V4 Flash amanai/deepseek-v4-flash 2.5 12.5 0.625
DeepSeek V4.1 Flash amanai/deepseek-v4.1-flash 2.5 12.5 0.625
DeepSeek V4 Flash 0731 amanai/deepseek-v4-flash-0731 2.5 12.5 0.625
MiniMax M3 amanai/minimax-m3 2.1 10.5 0.525
MiniMax M2.7 amanai/minimax-m2.7 1.2 6 0.3
Hunyuan 3 amanai/hy3 4.5 22.5 1.125
Hunyuan 4 amanai/hy4 4.5 22.5 1.125
Grok 4.5 amanai/grok-4.5 6 30 1.5
Grok 4.6 amanai/grok-4.6 6 30 1.5
GPT 5.5 amanai/gpt-5.5 10 50 2.5
GPT 6 Astra amanai/gpt-6-astra 30 150 7.5
GPT 5.6 Sol amanai/gpt-5.6-sol 10 50 2.5
GPT 5.6 Terra amanai/gpt-5.6-terra 7.5 37.5 1.875
GPT 5.6 Luna amanai/gpt-5.6-luna 5.625 28.125 1.40625
Claude Fable 5.1 amanai/claude-fable-5.1 30 150 7.5
Claude Fable 5 amanai/claude-fable-5 30 150 7.5
Claude Opus 5 amanai/claude-opus-5 15 75 3.75
Claude Opus 4.8 amanai/claude-opus-4.8 15 75 3.75
Claude Opus 4.7 amanai/claude-opus-4.7 15 75 3.75
Claude Opus 4.6 amanai/claude-opus-4.6 15 75 3.75
Claude Sonnet 5 amanai/claude-sonnet-5 10.5 52.5 2.625
Claude Sonnet 4.6 amanai/claude-sonnet-4.6 9 45 2.25
Claude Haiku 4.5 amanai/claude-haiku-4.5 3 15 0.75

Embedding model pricing

Token-metered embeddings: credits = max(250, ceil(tokens × model rate)). EmbeddingGemma remains byte-metered: credits = max(1,000, ceil(UTF-8 units × rate)). Local byte counts are not exact token counts.

ModelDimensionsRateMinimum
BGE Small EN amanai/bge-small-en 3841.25 credits / token250 credits
BGE Base EN amanai/bge-base-en 7684 credits / token250 credits
BGE Large EN amanai/bge-large-en 102412.5 credits / token250 credits
BGE-M3 amanai/bge-m3 10240.75 credits / token250 credits
EmbeddingGemma 300M amanai/embeddinggemma-300m 7681 credit / UTF-8 unit1,000 credits
Qwen3 Embedding 0.6B amanai/qwen3-embedding-0.6b 10240.75 credits / token250 credits
voyage-4-lite amanai/voyage-4-lite 10241.25 credits / token250 credits
voyage-4 amanai/voyage-4 10243.75 credits / token250 credits
voyage-4-large amanai/voyage-4-large 10247.5 credits / token250 credits
voyage-code-4 amanai/voyage-code-4 10247.5 credits / token250 credits
voyage-finance-2 amanai/voyage-finance-2 10247.5 credits / token250 credits
voyage-law-2 amanai/voyage-law-2 10247.5 credits / token250 credits

Rerank pricing

BGE and Voyage count tokens for the query and every submitted document. The rates and minimums below are model-specific; top_n never reduces billed input.

ModelRateMinimumCounted input
BGE Reranker Base amanai/bge-reranker-base 0.25 credits / token250 creditsQuery plus every submitted document; top_n does not reduce billed input
Voyage rerank-3 amanai/voyage-rerank-3 3 credits / token250 creditsQuery plus every submitted document; top_n does not reduce billed input
Voyage rerank-3-lite amanai/voyage-rerank-3-lite 1.25 credits / token250 creditsQuery plus every submitted document; top_n does not reduce billed input
Voyage rerank-2.5 amanai/voyage-rerank-2.5 3 credits / token250 creditsQuery plus every submitted document; top_n does not reduce billed input
Voyage rerank-2.5-lite amanai/voyage-rerank-2.5-lite 1.25 credits / token250 creditsQuery plus every submitted document; top_n does not reduce billed input

System One pricing

Jev returns structured decisions at 0 credits per request while the offer is active. Currently free for a limited time while the upstream free offering remains available. Free access is not guaranteed and future requests may be priced differently. Input and output token usage is still included in the response and request logs.

ModelVendorRateBilling
Jev 1.13 amanai/jev-1.13 TypeSafe AIFree · 0 creditsPer request while offer is active

Media model pricing

ModelCapabilityCurrent rateSettlement
GPT 5.6 Luna Image amanai/gpt-5.6-luna-image Image generation5,000,000 credits / delivered imageDelivered images only; unusable output is refunded
GPT 5.5 Image amanai/gpt-5.5-image Image generation7,500,000 credits / delivered imageDelivered images only; unusable output is refunded
GPT 5.6 Terra Image amanai/gpt-5.6-terra-image Image generation7,500,000 credits / delivered imageDelivered images only; unusable output is refunded
GPT 5.6 Sol Image amanai/gpt-5.6-sol-image Image generation10,000,000 credits / delivered imageDelivered images only; unusable output is refunded
Grok Imagine Image amanai/grok-imagine-image Image generation and editing3,000,000 credits / delivered imageDelivered images only; unusable output is refunded
Grok Imagine Image 2.0 amanai/grok-imagine-image-2.0 Image generation and editing4,000,000 credits / delivered imageDelivered images only; unusable output is refunded
Grok Imagine Video amanai/grok-imagine-video Video generation7,000,000 credits / second at 720pReserved at submission; settled or refunded once
Grok Imagine Video 1.5 amanai/grok-imagine-video-1.5 Video generation10,500,000 credits / second at 720pReserved at submission; settled or refunded once
Grok Voice TTS amanai/grok-voice-tts Speech synthesis2,000,000 credits / 1,000 characters · 500,000 minimumSettled after playable audio is delivered
Grok Voice Transcribe 1.0 amanai/grok-voice-transcribe-1.0 Speech transcription400,000 credits / minute · 250,000 minimumSettled after a transcript is delivered
Grok Voice Transcribe 2.0 amanai/grok-voice-transcribe-2.0 Speech transcription600,000 credits / minute · 250,000 minimumSettled after a transcript is delivered

Image output modifiers

SelectionMultiplierApplies to
quality=low0.75×GPT image models
quality=auto or medium1×GPT image models
quality=high1.5×GPT image models
1536x1024 or 1024x15361.5×GPT image models
Each additional edit input image+0.25×Where multi-image editing is supported
Grok Image 2.0 · low / 1K1×Generation default
Grok Image 2.0 · low / 2K or medium / 1K1.5×Output image
Grok Image 2.0 · medium / 2K2×Output image
Grok Image 2.0 edit source+0.25×Per accepted source image

For Grok Image 2.0, quality=auto is priced as low for generation and medium for editing. The modifier is based on the selected quality and resolution; aspect_ratio does not add a separate charge. The model's base price and the edit-source surcharge still apply.

Realtime pricing

ModelCurrent rateSettlement
GPT Live amanai/gpt-live 7,000,000 credits / successful sessionFailed session negotiation is not billed

Reserve and settlement

Before a costed call begins, the gateway atomically reserves a bounded worst-case amount. After delivery it settles the actual charge and returns any unused reserve.

✓A single request cannot spend beyond its reserve.
✓Concurrent requests cannot race past the available balance.
✓Failures before usable delivery are fully refunded; already delivered streaming usage may settle.
✓Durable jobs settle or refund exactly once.

Credit packs

Credits are prepaid and valid for 30 days. Account limits are shared across keys unless a key has a tighter scope.

PackCreditsQRISUSDCRPMConcurrency
Pro250,000,000Rp60,0004 USDC1208
Max500,000,000Rp90,0006 USDC20016
Ultra1,000,000,000Rp150,00010 USDC32024

Check the active balance with:

bash
curl https://api.amanai.dev/v1/usage -H "Authorization: Bearer $AMANAI_API_KEY"