Enterprise grade inference at scale. Custom model fine tuning you own.

Serverless inference
from openai import OpenAI
client = OpenAI(
    base_url="https://api.usbios.ai/v1", api_key="sk-bios-...",
)

resp = client.chat.completions.create(
    model="<model-id>",
    messages=[{"role": "user", "content": "Hello!"}],
    stream=True,
)
BiOS fine-tuning engine
Model
Base model
Method
SFT, RLHF, CPT
Adapter
LoRA, QLoRA, full
Dataset
Upload or import
GPU
Tier and budget
Backups
Fallback GPUs
Config
Hyperparameters
Review
Cost, then launch

Inference-First by Design

Inference is serverless by default. Scaling, traffic handling, and cost optimization happen automatically, including scaling to zero.

Serverless by Default

Inference runs serverless by default, with automatic scaling and request batching.

Performance at Scale

Dedicated GPU clusters ensure stable throughput under sustained load.

Flexible by Design

Scale from API-based inference to full GPU clusters without re-architecting your stack.

Custom models

Serve a model you trained, on a GPU of its own

Fine-tuning produces weights you own. A finished checkpoint can go straight onto a dedicated endpoint without leaving the platform.

Comes with fine-tuning

Custom Model Endpoints

Reached through the same OpenAI-compatible API as everything else. This is part of fine-tuning rather than the serverless product — different billing, and a different reason to reach for it.

Per second of GPU time
The same rates as training — from $0.42/hr across 12 GPU types

Adapters serve themselves

A LoRA or QLoRA adapter is served with its base model automatically, with no manual merge step.

The route matches the model

Chat, completion, embedding or reranker — each exposes the matching OpenAI route.

Sized to expected traffic

Light fits on the minimum that will hold the model; heavy adds GPUs for concurrency.

Memory is yours to tune

bf16 or fp8, with KV cache compression to fit more concurrent requests on the same card.

What it costs

See what it costs to run serverless inference on BiOS

BiOS saves you on deepseek-v4-pro

25%

vs Fireworks

25%

vs Together AI

Pick the model you use most

Choose a period, then a volume — or type your own

tokens/day

What share of your tokens are input rather than output

Input 70%Output 30%

Based on 1 billion tokens/day · 700 million input, 300 million output

Input / 1MOutput / 1MDaily
BiOS$1.30$2.60$1,690.00
Fireworks$1.74$3.48$2,262.00
Together AI$1.74$3.48$2,262.00

Over an average month BiOS is $51,404$17,398 saved against Fireworks, and $17,398 saved against Together AI.

Fireworks Standard tier, as published 26 July 2026 · Together AI serverless, as published 26 July 2026.

Start with $10 in credits

No credit card required. Billed per second of GPU time.

How you pay

Two products, two pricing models

Serverless inference is metered per million tokens. Fine-tuning and dedicated endpoints are metered per second of GPU time. You are never charged both ways for the same work.

Serverless inference
Per million tokens
  • Inputtokens you send, from $0.07 per 1M
  • Outputtokens generated, priced separately, from $0.26 per 1M
  • Prompt cachingbilled separately again

Rates vary by model — price yours in the calculator above. The rate for the exact model you are about to call is shown in your dashboard before you send the request.

Fine-tuning & dedicated endpoints
Per second of GPU time
  • Ratefrom $0.42/hr across 12 GPU types
  • Meteredby the second, for the time you actually run
  • Tokensnever counted, never charged

The same rate applies whether you are running SFT, RLHF, VLM or continued pre-training. A stopped endpoint costs nothing while it is stopped.

Neither product can run up a debt. Both draw from one pre-paid wallet. If the balance reaches zero you get a grace warning, and a running endpoint pauses rather than accruing charges. Top up and resume.

Which one should you use?

Neither is automatically cheaper. It depends on your traffic shape and whether you need your own model.

Reach for serverless inference
  • Spiky, bursty or unpredictable traffic
  • You want to ship today with no capacity decisions
  • Low or occasional volume, where an idle GPU would waste money
  • You are evaluating several models before committing to one
  • A stock production model already does the job
  • You would rather scale to zero than manage a floor
Reach for fine-tuning and a dedicated endpoint
  • The model has to know something only your data teaches it
  • Steady, high volume where per-token bills compound
  • Long prompts or large context, metered per token elsewhere
  • Compute that must not be shared with other tenants
  • You want a cost ceiling set by hardware, not by usage

We would rather say this plainly than have you discover it on an invoice. A dedicated GPU bills while it is idle, so stop endpoints you are not using.

Which one you need

When to fine-tune, and when not to

Fine-tuning costs more than a serverless call and takes longer to get right. Plenty of workloads should start on serverless and stay there. Here is how to tell which side you are on.

Stay on serverless inference

A general-purpose model is already doing the job — or you do not yet know exactly what the job is.

  • You are validating an idea or shipping a first version
  • Your prompts still change week to week
  • Traffic is spiky, seasonal, or low volume
  • An open model in the catalog already hits the accuracy you need
  • You have no labelled examples of the output you want

Fine-tune your own model

A general-purpose model gets close, but is consistently wrong in a way you can describe.

  • You have examples of the output you want — hundreds or thousands, not millions
  • You need a smaller, cheaper model to do one narrow job well
  • Your domain language is not in the base model — clinical, legal, internal jargon
  • Format or tone has to be right on every single call
  • You need the weights to belong to you

Three reasons not to fine-tune

To add facts the model should look up. Knowledge that changes belongs in retrieval, not in weights — bake it in and you retrain every time it moves. To rescue a prompt you have not iterated on. Prompt work costs minutes and nothing per attempt; training costs neither of those things. To win a public benchmark. A model that scores well on a leaderboard and badly on your own traffic has cost you money. Measure on your data, not on someone else's.

The API

Nothing to rewrite

The API is OpenAI-compatible. Point any OpenAI SDK at the base URL, pass your key as the bearer token, and set the model name. Switching models, or moving from serverless to your own dedicated endpoint, is a change of model id — not a rewrite. The serving engine covers four task shapes:

Model typeRouteBest forStreaming
Chat/v1/chat/completionsAssistants, multi-turn conversationYes
Completion/v1/completionsBase or completion-style modelsYes
Embedding/v1/embeddingsRetrieval and similarity searchNo
Reranker/v1/rerankCross-encoder document scoringNo

Tool calling that never breaks the contract

Tool support is read from the model's own chat template, not its name, so a custom fine-tune that speaks a known format gets it automatically. Once a tool request is accepted you always get an answer — if the output cannot be parsed as tool calls you get the raw text and an honest finish reason, never an error.

JSON that is always valid

Structured output works on every chat model, independent of tool support. Pass a response_format and the engine constrains decoding, so you get valid JSON — or an exact match to your schema when you supply one.

Dedicated endpoints

An endpoint you can actually operate

When you do run your own dedicated endpoint, you get the operational surface that decides whether you keep it: what it costs, how fast it is, and why it restarted at 3am.

Totals that never reset

Requests, prompt and generated tokens, cache-hit rate, average time-to-first-token, end-to-end latency, aborted requests, uptime and spend — cumulative forever, carried across engine and pod restarts.

Time-series metrics

Requests per minute, tokens per second in and out, TTFT and latency percentiles, running and queued requests, cache-hit rate, GPU utilisation and memory. 1h, 24h, 7d and 30d windows.

Self-healing, and honest about it

A crashed engine restarts with backoff. A crash loop is detected and the pod is stopped so it stops charging you, with the cause classified rather than hidden.

Playground built in

Chat with or prompt a running endpoint in the dashboard, with streaming, before you write any client code. It adapts to the model type and uses the endpoint key for you.

Visible startup

Provisioning, downloading weights, loading model, ready — with a byte-accurate download bar. The provisioning deadline scales with model size, so a large model is never killed for being slow to start.

Managed by API too

Create, stop, resume, restart, reconfigure and rotate keys over a REST management API, with idempotency keys on create so a timeout never leaves you with two endpoints.

Control when you want it

Sensible defaults. Real levers underneath.

Every serving field has a working default — you can deploy without touching any of them. When you do need to tune for cost or throughput, the levers are real, and the platform tells you what each one costs you in GPU memory before you commit.

A live VRAM bar shows the weights / KV-cache / overhead split as you change context length, quantization or GPU count. The minimum count that fits is enforced, the recommended count is starred, and cards that cannot run your chosen precision are disabled with the reason.

Model typeChat, completion, embedding or reranker — sets the route exposed
Expected trafficLight, steady or heavy — drives the recommended GPU count
Context lengthDefaults to min(native, 32K), hard-capped at the model native max
Quantizationbf16 full precision, or fp8 for roughly half the memory
KV cacheauto, FP8, or TurboQuant 4/3/2-bit to fit more concurrent requests
GPU type & countLive VRAM fit and hourly price, up to 8 GPUs in a single pod

Product 02 — Fine-tuning

Train a model that knows your domain

A separate product with its own pricing. Bring your data, pick a method and an algorithm, and train on dedicated GPUs billed by the second.

15+ fine-tuning methods

LoRA, QLoRA, AdaLoRA, LoHa, LoKr, BOFT, OFT, VeRA, ReFT and full parameter training.

6 alignment algorithms

DPO, SimPO, CPO, ORPO, KTO and Reward Modeling, without building an RL pipeline.

120+ vision-language models

InternVL, Qwen-VL, LLaVA, DeepSeek-VL, MiniCPM-V, Phi Vision and more.

Continued pre-training

Extend a foundation model with proprietary data — medical, legal, financial, code.

QwenLlamaDeepSeekPhiGemmaMistralInternVLLLaVAMixtralChatGLM

The weights your jobs produce are yours

This is not a positioning statement, it is in the contract. Our Terms state that the datasets you upload and the fine-tuned model weights your jobs produce are yours, and that we claim no rights over them beyond storing and processing them on your behalf. Base models keep their own licences, and complying with those remains your responsibility.

Start with either one

Call a production model through the serverless API, or train your own on dedicated GPUs. Two products, one account, no commitment on either.

Serverless per million tokens · Fine-tuning from $0.42/hr · Your weights stay yours