How does SpeedInfer token metering and billing work?

Inference is billed authoritatively per token based on exact upstream consumption. A conservative credit reservation is held when the request starts, and unused balance is refunded immediately when the response completes.

Are there monthly platform fees or minimum spending commitments?

No. The Pay-As-You-Go API requires no monthly subscription or upfront contract. You only pay for the tokens you generate with prepaid account credits.

Can I use standard OpenAI SDKs with SpeedInfer?

Yes. SpeedInfer is a 100% drop-in replacement for OpenAI API endpoints. Simply point base_url to https://speedinfer.com/v1 and provide your SpeedInfer API key.

What is the difference between Serverless and Dedicated Endpoints?

Serverless inference routes requests on-demand across shared connected model workers and external providers with zero idle charges. Dedicated Endpoints provide isolated single-tenant model instances with guaranteed concurrency and custom fine-tuned LoRA checkpoints for enterprise production workloads.

TRANSPARENT & METERED PRICING

Predictable compute.
Zero idle waste.

Pay solely for the tokens you stream across top open foundation models, or plan dedicated worker endpoints with dedicated orchestration.

Developer Choice

SERVERLESS INFERENCE

Pay-As-You-Go API

Instant access via OpenAI SDK. High availability with zero monthly commitments.

$0.20 / 1M prompt tokens

$0.60 / 1M completion tokens

✓ Drop-in OpenAI API SDK compatibility
✓ Real-time Server-Sent Events (SSE) streaming
✓ Dual-source routing: Connected workers & closed providers
✓ Authoritative exact token metering & refunds
✓ Scoped API keys with per-key RPM & TPM limits
✓ No minimum commitment or monthly fee

Start Free Trial ↗

Enterprise Tier

PROVISIONED CAPACITY

Dedicated Endpoints

Isolated single-tenant model serving, custom LoRA weights, and guaranteed provisioned throughput.

Custom / tailored concurrency

Tailored to your production volume and SLA requirements

✓ Single-tenant model isolation for zero-queue latency
✓ Custom LoRA & QLoRA fine-tuned checkpoint deployment
✓ Dynamic tensor parallelism & FP8 acceleration
✓ Private VPC networking & custom security boundaries
✓ S3-compatible dataset & weights bucket storage
✓ Enterprise telemetry and custom SLA agreements

Talk to Sales ↗

TOKEN PRICING BREAKDOWN

Foundation Models Catalog Pricing

Prices include OpenRouter upstream base + 5% platform markup

Model	Type	Context Window	Input (per 1M)	Output (per 1M)	Execution Layer
DeepSeek V3 `deepseek/deepseek-chat`	Open Weights	163,840	$0.2703	$1.0801	Connected NVIDIA GPU Worker
DeepSeek R1 `deepseek/deepseek-r1`	Open Weights	64,000	$0.7350	$2.6250	Connected NVIDIA GPU Worker
Qwen3 235B A22B `qwen/qwen3-235b-a22b`	Open Weights	131,072	$0.4777	$1.9110	Connected NVIDIA GPU Worker
Qwen3 30B A3B `qwen/qwen3-30b-a3b`	Open Weights	131,072	$0.1365	$0.5460	Connected NVIDIA GPU Worker
Qwen 2.5 72B Instruct `qwen/qwen-2.5-72b-instruct`	Open Weights	32,768	$0.3780	$0.4200	Connected NVIDIA GPU Worker
Qwen 2.5 Coder 32B `qwen/qwen-2.5-coder-32b-instruct`	Open Weights	32,768	$0.6930	$1.0500	Connected NVIDIA GPU Worker
Qwen 2.5 7B Instruct `qwen/qwen-2.5-7b-instruct`	Open Weights	32,768	$0.1050	$0.2100	Connected NVIDIA GPU Worker
Llama 4 Maverick `meta-llama/llama-4-maverick`	Open Weights	1,048,576	$0.1969	$0.6851	Connected NVIDIA GPU Worker
Llama 4 Scout `meta-llama/llama-4-scout`	Open Weights	1,310,720	$0.1050	$0.3150	Connected NVIDIA GPU Worker
Llama 3.3 70B Instruct `meta-llama/llama-3.3-70b-instruct`	Open Weights	131,072	$0.1050	$0.3360	Connected NVIDIA GPU Worker
Hermes 3 (Llama 3.1 405B) `nousresearch/hermes-3-llama-3.1-405b`	Open Weights	131,072	$1.0500	$1.0500	Connected NVIDIA GPU Worker
Llama 3.1 70B Instruct `meta-llama/llama-3.1-70b-instruct`	Open Weights	131,072	$0.4200	$0.4200	Connected NVIDIA GPU Worker
Llama 3.1 8B Instruct `meta-llama/llama-3.1-8b-instruct`	Open Weights	131,072	$0.0525	$0.0840	Connected NVIDIA GPU Worker
Llama 3.2 3B Instruct `meta-llama/llama-3.2-3b-instruct`	Open Weights	131,072	$0.0525	$0.3465	Connected NVIDIA GPU Worker
Llama 3.2 1B Instruct `meta-llama/llama-3.2-1b-instruct`	Open Weights	60,000	$0.0284	$0.2111	Connected NVIDIA GPU Worker
NVIDIA Nemotron 3 Ultra `nvidia/nemotron-3-ultra-550b-a55b`	Open Weights	262,144	$0.6300	$2.5200	TensorRT-LLM Connected Worker
Nemotron 3 Super 120B `nvidia/nemotron-3-super-120b-a12b`	Open Weights	262,144	$0.0840	$0.4725	TensorRT-LLM Connected Worker
Mistral Small 3.1 24B `mistralai/mistral-small-3.1-24b-instruct`	Open Weights	128,000	$0.3685	$0.5827	Connected NVIDIA GPU Worker
Mistral Large `mistralai/mistral-large`	Open Weights	128,000	$2.1000	$6.3000	Connected NVIDIA GPU Worker
Codestral 2508 `mistralai/codestral-2508`	Open Weights	256,000	$0.3150	$0.9450	Connected NVIDIA GPU Worker
Mistral NeMo `mistralai/mistral-nemo`	Open Weights	131,072	$0.0200	$0.0315	Connected NVIDIA GPU Worker
Google Gemma 3 27B `google/gemma-3-27b-it`	Open Weights	131,072	$0.0840	$0.4725	Connected NVIDIA GPU Worker
Google Gemma 3 12B `google/gemma-3-12b-it`	Open Weights	131,072	$0.0525	$0.1575	Connected NVIDIA GPU Worker
Google Gemma 3 4B `google/gemma-3-4b-it`	Open Weights	131,072	$0.0525	$0.1050	Connected NVIDIA GPU Worker
Microsoft Phi-4 `microsoft/phi-4`	Open Weights	16,384	$0.0735	$0.1470	Connected NVIDIA GPU Worker
GPT-5.6 Sol `openai/gpt-5.6-sol`	Proprietary	1,050,000	$2.1000	$10.5000	External Provider Routing
GPT-5.6 Terra `openai/gpt-5.6-terra`	Proprietary	1,050,000	$2.1000	$12.6000	External Provider Routing
GPT-5.6 Luna `openai/gpt-5.6-luna`	Proprietary	1,050,000	$0.2100	$1.2600	External Provider Routing
OpenAI o1 `openai/o1`	Proprietary	200,000	$15.7500	$63.0000	External Provider Routing
OpenAI o3-mini `openai/o3-mini`	Proprietary	200,000	$1.1550	$4.6200	External Provider Routing
OpenAI GPT-4o `openai/gpt-4o`	Proprietary	128,000	$2.6250	$10.5000	External Provider Routing
Claude Fable 5 `anthropic/claude-fable-5`	Proprietary	1,000,000	$10.5000	$52.5000	External Provider Routing
Claude Opus 5 `anthropic/claude-opus-5`	Proprietary	1,000,000	$5.2500	$26.2500	External Provider Routing
Claude Sonnet 5 `anthropic/claude-sonnet-5`	Proprietary	1,000,000	$2.1000	$10.5000	External Provider Routing
Claude 3.5 Sonnet `anthropic/claude-3.5-sonnet`	Proprietary	200,000	$3.1500	$15.7500	External Provider Routing
Claude Haiku 4.5 `anthropic/claude-haiku-4.5`	Proprietary	200,000	$1.0500	$5.2500	External Provider Routing

FREQUENTLY ASKED QUESTIONS

Everything you need to know about billing.

How does SpeedInfer handle credit reservations?

Before dispatching inference to a worker, the gateway reserves a conservative balance based on your requested max_tokens. When the response finishes, SpeedInfer authoritatively settles against the exact tokens reported by the runtime engine and immediately refunds unconsumed credits to your balance.

What payment methods are supported?

Prepaid credit top-ups are processed securely via certified PCI-DSS compliant payment gateways (supporting credit cards, debit cards, and Apple Pay). Enterprise custom invoicing with wire transfers is available for monthly volumes over $2,000.

Are failed or interrupted requests billed?

No. If a model worker experiences an unrecoverable failure or disconnects mid-stream, the reservation is released and your balance is not charged for unserved completions.

Can I host my fine-tuned LoRA checkpoints?

Yes. Custom LoRA adapters fine-tuned on SpeedInfer or uploaded through S3-compatible buckets can be loaded onto Dedicated Endpoints or hot-swapped for low-latency specialized inference.

Ready to build with SpeedInfer?

Start Building ↗ Read API Docs ↗

Predictable compute.Zero idle waste.