HuggingFace mirror: — Browse all on HF ↗

Inference API ✕

Streaming

Fine-tunable

Quantized

Safetensors

>10B params

Chat template

Tool use

Vision

Quick APIFull docs →

POST/v1/meta-llama/Llama-3-8B/generate

// stream inference fetch("https://hf.run.place/v1/{model}", { method: "POST", headers:{"X-API-Key": key}, body: JSON.stringify({ inputs: prompt, stream: true, max_new_tokens: 512 }) })

Live Stream Log status ↗

Platform statsStatus ↗

—

req / min

—

tokens / s

99.97%

uptime 90d

47ms

avg latency

1,847,293

total requests

9.28B

total tokens

61

models live

Top request originsAPI usage →

🇺🇸 United States		78%
🇨🇳 China		52%
🇩🇪 Germany		31%
🇬🇧 United Kingdom		24%
🇷🇺 Russia		18%

Stream Tester

Test real-time token streaming from any model endpoint.

Model

Max tokens

Temperature

Top-p

Prompt

Ready

// Output will appear here token by token…

Model Comparison

Side-by-side benchmark scores. Click any row to open on HuggingFace.

Model	Params	MMLU	HumanEval	MT-Bench	Context	License
DeepSeek-R1	671B	90.8	84.4	9.38	128k	MIT
Qwen2.5-72B-Instruct	72B	86.1	74.3	9.12	128k	Apache 2.0
Llama-3.1-70B-Instruct	70B	83.6	72.8	8.91	128k	Llama
Llama-3.3-70B-Instruct	70B	85.1	74.9	9.05	128k	Llama
DeepSeek-R1-Distill-32B	32B	87.2	78.1	9.21	128k	MIT
Mixtral-8x22B-Instruct	141B	77.8	59.1	8.66	64k	Apache 2.0
Phi-4 (14B)	14B	84.8	82.6	8.92	16k	MIT
Qwen2.5-Coder-32B-Instruct	32B	81.3	88.2	8.74	128k	Apache 2.0
Gemma-2-27B-it	27B	75.2	51.8	7.88	8k	Gemma
Gemma-2-9B-it	9B	71.3	46.4	7.64	8k	Gemma
Mistral-7B-v0.3	7B	64.2	41.7	7.41	32k	Apache 2.0
Mistral-Nemo-12B	12B	68.0	45.2	7.62	128k	Apache 2.0
Phi-3.5-mini-instruct	3.8B	69.0	58.3	8.02	128k	MIT
Llama-3.1-8B-Instruct	8B	73.0	61.3	8.11	128k	Llama
Qwen2.5-7B-Instruct	7B	74.2	62.5	8.19	128k	Apache 2.0
Mixtral-8x7B-Instruct	47B	70.6	56.4	8.30	32k	Apache 2.0

Streaming ✕

Viewer available

Croissant

CC licensed

No PII

📚

Open-Hermes-2.5

900k high-quality instruction samples across diverse tasks.

⬇ 2.1M · ♥ 8.4k · Apache 2.0

💬

ShareGPT4

68k GPT-4 conversations covering coding, math, reasoning and general chat.

⬇ 1.8M · ♥ 6.1k · CC BY 4.0

🧮

MetaMathQA

395k math reasoning problems with step-by-step solutions from GSM8K and MATH.

⬇ 980k · ♥ 3.2k · MIT

🖼️

LAION-5B

5.85 billion image-text pairs. The largest openly available image-text dataset.

⬇ 740k · ♥ 12k · CC BY 4.0

🌍

CC-100

Web crawl data across 100+ languages for multilingual pretraining.

⬇ 620k · ♥ 4.8k · CC BY 4.0

💻

The-Stack-v2

900B tokens of permissively licensed source code in 600+ languages.

⬇ 580k · ♥ 9.1k · Various

Dataset Leaderboard

Ranked by downloads in the last 30 days. Click any row to open on HuggingFace.

#	Dataset	Org	Task	Downloads	Likes	License
1	OpenHermes-2.5	teknium	Instruct	2.1M	8.4k	Apache 2.0
2	LAION-5B	laion	Image-Text	740k	12k	CC BY 4.0
3	The-Stack-v2	bigcode	Code	580k	9.1k	Various
4	ShareGPT4	shibing624	Chat	1.8M	6.1k	CC BY 4.0
5	MetaMathQA	meta-math	Math QA	980k	3.2k	MIT
6	CC-100	statmt	Pretraining	620k	4.8k	CC BY 4.0

🤖

chat-with-llama3

Interactive chatbot powered by Llama 3.1 70B. Multi-turn dialogue with system prompts.

👁 240k views · Gradio · A10G

🎨

FLUX-playground

Text-to-image with FLUX.1-dev. Exceptional photorealism and prompt adherence.

👁 580k views · Gradio · A100

🎙️

whisper-webui

Drag-and-drop audio transcription. Whisper Large v3. Supports 99 languages.

👁 190k views · Gradio · T4

📊

LLM Leaderboard

Open benchmark leaderboard tracking 1,200+ models across 6 eval suites.

👁 1.2M views · Streamlit · CPU

🧬

ESMFold-protein

Single-sequence protein structure prediction with 3D viewer.

👁 88k views · Gradio · A10G

🎬

AnimateDiff-video

Text-to-video with AnimateDiff. 16-frame 512px clips in under 30s.

👁 310k views · Gradio · A100

Getting Started

API Reference

Guides

Introduction

huggingface.run.place provides a unified REST API for inference on 430,000+ models from the HuggingFace Hub. Compatible with the OpenAI API schema — swap base_url and you're done.

All endpoints support real-time token streaming via Server-Sent Events (SSE). Response latency is typically under 50ms to first token.

Base URL

ENDPOINT

https://huggingface.run.place/v1

Quick start — Python

PYTHON

from openai import OpenAI
client = OpenAI(
    base_url="https://huggingface.run.place/v1",
    api_key="hfrp_..."
)
stream = client.chat.completions.create(
    model="meta-llama/Llama-3.1-70B-Instruct",
    messages=[{"role":"user","content":"Hello!"}],
    stream=True
)
for chunk in stream:
    print(chunk.choices[0].delta.content, end="")

External resources

🤗

HF Transformers Docs

huggingface.co/docs/transformers

↗ 🔮

OpenAI API Reference (compatible)

platform.openai.com/docs/api-reference

↗ 🦙

llama.cpp

github.com/ggerganov/llama.cpp

↗

API Explorer

Interact with the live API directly from your browser. Requests are counted against your quota.

Request Body

Response

// Click Send to run…

Endpoint	Method	Description	Stream
/v1/generate	POST	Raw text generation	✓
/v1/chat/completions	POST	OpenAI-compatible chat	✓
/v1/embeddings	POST	Dense vector embeddings	—
/v1/images/generate	POST	Text-to-image (SDXL, FLUX)	—
/v1/audio/transcribe	POST	Speech-to-text (Whisper)	✓
/v1/models	GET	List available models	—

Simple, token-based pricing

Pay only for what you use. No seat fees, no egress costs, no hidden charges.

Starter

^$0

Free forever

10M tokens / month
All open-source models
3 API keys
Community support
Streaming enabled

Token pricing

All models billed per token. No minimums, no monthly commitments on pay-as-you-go.

Model	Type	Input / 1M tok	Output / 1M tok	Context
DeepSeek R1 671B ↗	MoE	$0.55	$2.19	128k
DeepSeek R1 Distill 32B ↗	LLM	$0.18	$0.54	128k
Llama 3.3 70B ↗	LLM	$0.35	$0.60	128k
Llama 3.1 70B ↗	LLM	$0.35	$0.60	128k
Llama 3.1 8B ↗	SLM	$0.05	$0.08	128k
Qwen 2.5 72B ↗	LLM	$0.35	$0.60	128k
Qwen 2.5 7B ↗	SLM	$0.07	$0.12	128k
Qwen 2.5 Coder 32B ↗	LLM	$0.18	$0.35	128k
Mixtral 8x22B ↗	MoE	$0.45	$0.70	64k
Mixtral 8x7B ↗	MoE	$0.24	$0.24	32k
Mistral Nemo 12B ↗	LLM	$0.13	$0.13	128k
Phi-4 14B ↗	LLM	$0.14	$0.14	16k
Phi 3.5 mini ↗	SLM	$0.10	$0.15	128k
Gemma 2 27B ↗	LLM	$0.27	$0.45	8k
Gemma 2 9B ↗	SLM	$0.09	$0.09	8k
BGE-M3 / E5 embed ↗	Embed	$0.02	—	8k
Whisper large-v3 ↗	ASR	$0.006 / min audio		30 min
Whisper large-v3-turbo ↗	ASR	$0.003 / min audio		30 min
FLUX.1-dev ↗	Image	$0.04 / image (1MP)		—
FLUX.1-schnell ↗	Image	$0.01 / image (1MP)		—
SD3.5 Large ↗	Image	$0.065 / image (1MP)		—

All prices in USD. Billed monthly. Pro plan includes a base of 200M tokens before per-token billing kicks in.

Enterprise

Dedicated capacity, SLAs, and hands-on support for teams building at scale.

🔒

Private deployments

Run models inside your own VPC or on-premise. No data leaves your environment.

⚡

Reserved capacity

Guaranteed GPU allocation with zero cold-start latency. Autoscales within your reserved pool.

🛡️

99.99% SLA

Contractual uptime commitment backed by 24/7 on-call engineering support.

🔑

SSO / SAML

Integrate with Okta, Azure AD, Google Workspace. Fine-grained RBAC per team.

📊

Custom analytics

Real-time usage dashboards, cost attribution by team/project, anomaly alerts.

🤝

Dedicated CSM

Named customer success manager, quarterly reviews, model recommendation service.

Talk to our team

We'll put together a custom proposal based on your workload, preferred models, and compliance requirements. Most enterprise deals close in under a week.

Blog

Updates, tutorials and deep-dives from the run.place team.

● All systems operational Updated just now

✅

All Systems Operational

Last checked — · 99.97% uptime over 90 days

Inference API● Operational

90 days ago99.98% uptimeToday

Model Hub / CDN● Operational

90 days ago99.99% uptimeToday

Authentication / API Keys● Operational

90 days ago100% uptimeToday

Dashboard / Web UI● Operational

90 days ago99.97% uptimeToday

API Response Time (last 24h)p50 · p95 · p99

■ p50: 47ms ■ p95: 148ms ■ p99: 312ms

Incident History

Degraded performance — Inference API (EU region) Jan 14, 2025 · Resolved

A network partition in the Frankfurt datacenter caused elevated latencies (p99 ~1.2s) for approximately 22 minutes. US and APAC regions were unaffected. Root cause: misconfigured load balancer rule during routine maintenance. Deployed fix at 14:22 UTC.

Duration: 22 min · Impact: <3% of requests · Severity: minor

Scheduled maintenance — Model weight sync Dec 29, 2024 · Completed

Routine maintenance to sync new model weights for DeepSeek-R1 and FLUX.1-dev. No user-facing impact. Completed ahead of schedule at 03:41 UTC.

Duration: 41 min · Impact: none · Severity: none

Authentication service slow responses Nov 3, 2024 · Resolved

Redis cache eviction caused token validation latency to spike to ~400ms for 8 minutes. API keys continued to work; only first-auth flow was affected. Cache TTL tuned.

Duration: 8 min · Impact: <1% of requests · Severity: minor

Get incident notifications

Subscribe to receive email or Slack alerts when incidents are opened or resolved.

Changelog

Every update, release, and improvement — in order.

Jan 28, 2025

NEW MODELDeepSeek-R1 671B now available

DeepSeek-R1, the MIT-licensed 671B mixture-of-experts reasoning model, is now fully supported. Chain-of-thought streaming works out of the box. SOTA on AIME 2024, MATH-500, and SWE-bench Verified.

View model →

Jan 14, 2025

INFRAp50 latency reduced to 47ms

Speculative decoding and continuous batching improvements cut median first-token latency from 80ms → 47ms globally. EU region saw the biggest improvement (−52%).

Jan 9, 2025

FEATURETool calling for Llama 3.3 70B

Llama 3.3 70B now supports parallel function calling with up to 8 tools per request. JSON schema validation built-in.

Dec 18, 2024

NEW MODELPhi-4 14B added

Microsoft Phi-4 is now available. SOTA on GPQA Diamond and MATH at 14B parameters.

Dec 5, 2024

APIBatch inference endpoint GA

POST /v1/batch is now generally available. Send up to 100 requests in one call. 30% cost reduction vs. individual requests at high volume.

Nov 20, 2024

NEW MODELQwen 2.5 family — 7B, 72B, Coder 32B

Qwen 2.5 instruct models now live. Qwen2.5-Coder-32B is the new SOTA for code completion on HumanEval.

Nov 1, 2024

FEATUREAPAC region launched (Tokyo)

New inference cluster in Tokyo (ap-northeast-1). APAC users see ~60% latency reduction. Automatic region routing based on IP.

Oct 14, 2024

INFRAH100 cluster upgrade complete

All 70B+ models now served from H100 SXM5 nodes. Throughput improved 2.4× for Llama 3.1 70B and Mixtral 8x22B.

Sep 27, 2024

LAUNCHPublic beta launched 🎉

huggingface.run.place opens to the public. 10 models available at launch, free tier with 1M tokens/month. 2,400 developers signed up in the first 48 hours.

org / Model Name

Description

About this model

Extended description loads here.

Intended use

Use case description.

Stats

Pricing

Reviews

Model family	Input	Output	Context
Llama 3.1 70B ↗	$0.35 / 1M	$0.60 / 1M	128k
Qwen 2.5 72B ↗	$0.35 / 1M	$0.60 / 1M	128k
Mixtral 8x22B ↗	$0.45 / 1M	$0.70 / 1M	64k
DeepSeek R1 671B ↗	$0.55 / 1M	$2.19 / 1M	128k
Phi 3.5 mini ↗	$0.10 / 1M	$0.15 / 1M	128k
Embeddings (BGE, E5) ↗	$0.02 / 1M	—	8k
Whisper large-v3 ↗	$0.006 / min	—	30 min