
Two of the largest corporate spend platforms in the US have independently named Vast.ai one of their fastest-growing vendors — not based on fundraising announcements or press coverage, but on what companies are actually spending money on.

**[Ramp](https://ramp.com/velocity/top-saas-vendors-on-ramp-march-2026)**, which tracks spend across 50,000+ businesses, named Vast.ai a breakout growth vendor in their March 2026 Velocity report. They categorized us under "Agent hosting" and noted that half the trending vendors on the platform are compute infrastructure providers. Their analysis:

> "AI agent workloads are moving from prototypes to production systems, and the constraint is shifting from 'which model is best' to 'where can I run this reliably.'"

Three months earlier, **[Brex](https://www.brex.com/journal/brex-benchmark-december-2025)** ranked Vast.ai **#11** in their 50 fastest-growing software vendors of 2025, based on transaction data from 35,000+ customers. [Fortune covered the report](https://fortune.com/2025/12/23/as-ai-investors-fret-over-roi-these-startups-attracted-serious-cash-customers-2025/), highlighting Vast.ai as a key infrastructure-layer company drawing real dollars alongside Groq (#12) and Supabase (#23).

That's 85,000+ businesses across two independent platforms telling the same story.

### What's driving the growth

The numbers behind these rankings reflect a shift we've been seeing on our own platform. Over the past twelve months:

- **New signups grew 27x year-over-year.** In February 2026 alone, more than 19,000 new users signed up for Vast.ai — compared to roughly 700 in February 2025.
- **Monthly active paying customers grew 16x**, from under 1,000 to over 14,000.
- **GPU supply scaled to meet demand.** Our marketplace now has nearly 20,000 GPUs across 5,000 machines from over 1,000 independent hosts — up 48% year-over-year.
- **Revenue grew 13x**, crossing seven figures per month for the first time.

This isn't growth driven by a single large customer or a one-time event. It's thousands of teams — researchers, startups, and enterprises — independently discovering that they can get GPU compute faster, cheaper, and with more flexibility on Vast.ai than anywhere else.

### Why corporate spend data matters

Fundraising rounds and press mentions are leading indicators. Corporate spend is a lagging one — it only moves after teams have evaluated, tested, and committed. When Ramp and Brex both flag the same company, it means real budgets are shifting. Not pilot programs. Not free-tier experiments. Paid production workloads.

Ramp's observation that the bottleneck is moving from "which model" to "where do I run it" is exactly the opportunity we built Vast.ai for. The platform is model-agnostic, cloud-agnostic, and contract-free. Teams deploy in seconds and pay by the second. That simplicity compounds — once a team starts, their usage tends to grow.

### What's next

We're investing in the areas that matter most to production teams: reliability, autoscaling, and deeper integrations for agentic workloads. The infrastructure layer for AI is still being built, and we plan to keep earning our position in these rankings month over month.

Thank you to the 100,000+ users who've signed up for Vast.ai, and to the 1,000+ hosts whose GPUs make this marketplace possible.

One more thing: we're also a Ramp customer ourselves. We use Ramp to manage our own corporate spend. Vendor and customer — it's a good sign when the tools you rely on are also relying on you.

---

*Links:*
- *[Ramp Velocity: Top SaaS Vendors — March 2026](https://ramp.com/velocity/top-saas-vendors-on-ramp-march-2026)*
- *[Brex: 50 Fastest-Growing Software Vendors of 2025](https://www.brex.com/journal/brex-benchmark-december-2025)*
- *[Fortune: AI Startups Attracting Serious Customer Cash in 2025](https://fortune.com/2025/12/23/as-ai-investors-fret-over-roi-these-startups-attracted-serious-cash-customers-2025/)*


Two of the largest corporate spend platforms independently ranked Vast.ai among their fastest-growing vendors — based on real transaction data from 85,000+ businesses.

Vast.ai Named Among Fastest Growing Vendors by Ramp and Brex


Welcome back to another round of updates at Vast.ai. We've kicked off the new year with improvements focused on expanding orchestration support, refining billing and serverless workflows, and introducing new templates and guides to help you deploy faster and more securely.

Another exciting update: we recently launched the [Vast.ai Startup Program](https://vast.ai/startup), designed to support early-stage teams building on our platform.

Startups can now get up to **$2,500 in free GPU credits** when they move their workloads to Vast.ai. With 24/7 priority support and access to Vast's full range of GPUs, the program delivers high-performance compute to power the next wave of AI innovation without long-term commitments. For more info, see our blog post [here](https://vast.ai/article/vast-ai-startup-program).

Not to mention we have a comprehensive new [Model Library](https://vast.ai/model-library), giving you a faster path from model selection to deployment.

## NVIDIA Cloud GPU Updates

### SkyPilot Enhancements

For teams using [SkyPilot](https://vast.ai/article/vast-ai-gpus-can-now-be-rentend-through-skypilot) to manage deployments, we've made updates that expand configuration options on Vast.ai. SkyPilot users can now filter capacity by Secure Cloud to ensure that workloads run only on verified datacenter-hosted GPUs.

We've also added full support for instance creation parameters via SkyPilot, enabling precise control over images, environment variables, disk size, pricing, startup commands, and other instance-level settings at launch. See [SkyPilot docs](https://docs.skypilot.co/en/latest/reference/config.html#vast) for full details.

### Fresh New Templates and Guides

Our [Model Library](https://vast.ai/model-library) introduces a model-first approach to deploying AI. Templates are organized by modality and use case and include recommended hardware presets and clear dedicated GPU pricing.

Here are some of our newest templates:

- [Kimi K2.5](https://vast.ai/model/kimi-k2.5) - An open-source, native multimodal agentic model that combines vision and language understanding with agentic capabilities.
- [LTX-2](https://vast.ai/model/ltx-2) - A DiT-based audio-video generation model that produces synchronized video and sound.
- [SGLang](https://cloud.vast.ai/template/readme/1fce7860e11071cae9c357fb1c8b7560) - A high-performance LLM inference framework with a hosted API server template.

And our latest guides:

- [dstack + vLLM](https://docs.vast.ai/examples/serving-infrastructure/dstack-vllm): Deploy language models using vLLM with dstack for automated provisioning and cost control.
- [DR-Tulu](https://docs.vast.ai/examples/mcp/dr-tulu): Deploy a research agent trained with an MCP server and web tools.
- [GLiNER2](https://docs.vast.ai/examples/ner/gliner2): Run GLiNER2 for named entity recognition with custom entity types.
- [BrowseSafe](https://docs.vast.ai/examples/ai-agents/browsesafe): Deploy BrowseSafe to detect prompt injection attacks in HTML.
- [Overnight Ralph Loop](https://docs.vast.ai/examples/ai-agents/overnight-ralph-loop): Budget-friendly alternative workflow to Claude Code.

### Other Improvements

If you missed it late last year, we also launched [Vast.ai Serverless](https://vast.ai/article/vast-ai-serverless-automated-gpu-scaling). It enables automatic, pay-per-use scaling for AI workloads through a fully serverless API on Vast's globally distributed GPU cloud.

## Our Commitment

With new SkyPilot enhancements, startup opportunities, a model-first deployment experience, and improved inference workflows, we're continuing to build a platform that gives you more control with less friction.

Need help? Contact our customer support team anytime at [support@vast.ai](mailto:support@vast.ai) or join our [Discord](https://discord.gg/vast) server for tips, real-time updates, and ongoing community discussion.

## Change Log

### SkyPilot Enhancements

- Advanced configuration options for SkyPilot on Vast.ai.
- Support for filtering by Secure Cloud (`datacenter_only`).
- Full support for instance creation parameters via SkyPilot.
- See [SkyPilot docs](https://docs.skypilot.co/en/latest/reference/config.html#vast).

### New Templates

- [Kimi K2.5](https://vast.ai/model/kimi-k2.5): Multimodal agentic model.
- [LTX-2](https://vast.ai/model/ltx-2): Video and audio generation model.
- [SGLang](https://cloud.vast.ai/template/readme/1fce7860e11071cae9c357fb1c8b7560): High-performance LLM inference engine.
- Find them in the [Model Library](https://vast.ai/model-library).

### Template Updates

- Base Image and PyTorch now support **CUDA 13.1** and updated NVML libraries.
- Ostris AI Toolkit updates.
- ComfyUI updates.
- Llama.cpp (ssl) updates.
- SGLang onstart script fix.
- Pinokio v6.x updates.

### New Guides

- [dstack](https://docs.vast.ai/examples/serving-infrastructure/dstack-vllm): Serving infrastructure.
- [DR-Tulu](https://docs.vast.ai/examples/mcp/dr-tulu): MCP.
- [GLiNER2](https://docs.vast.ai/examples/ner/gliner2): Named Entity Recognition.
- [BrowseSafe](https://docs.vast.ai/examples/ai-agents/browsesafe): AI agents and security.
- [Overnight Ralph Loop](https://docs.vast.ai/examples/ai-agents/overnight-ralph-loop): Budget-friendly alternative to Claude Code.

### Platform Improvements

- Fixed serverless GPU filter UI.
- Fixed serverless template selection UI.
- Billing Transaction History Invoice PDF Export now excludes failed and pending payments and other non-payment activity.
- Billing Transaction History now supports easy individual invoice download.
- Security improvements.
- Local volume deletion issue resolved.
- Copy progress bar issue resolved.
- Disk quota issues resolved.
- Machines missing geolocation resolved.
- Host CLI schedule maintenance command issue resolved.


Vast.ai's January and February 2026 updates include new SkyPilot controls, Startup Program credits, Model Library additions, fresh guides, and major platform fixes.

January & February 2026 Product Updates


When you think of a large language model (LLM), you probably picture the big household names like ChatGPT, Claude, Gemini, and Llama. They're dominant players for a reason – delivering cutting-edge performance that sets the standard for general-purpose AI – but they're also overwhelmingly centered around the English language and Western culture.

If you look further, there are plenty of other models gaining traction globally. Fueled by government investment as well as tech giants, startups, and research labs, these regional LLMs are trained on local languages and optimized for diverse needs.

With that in mind, let's take a look at some examples from around the world.

### Middle East: Arabic-First AI

Across the Middle East, government partnerships and large-scale investments are driving the rapid expansion of the region's AI sector. Much of this effort is focused on Arabic-first and bilingual LLMs designed for regional sovereignty and cultural awareness. For instance, the following are some of the leading projects in the region:

*   Jais offers open bilingual models up to 30B parameters, trained on large Arabic-English datasets, with the ability to switch between Arabic dialects. Instruction-tuned "Chat" variants, optimized for conversational applications, have also been released.

*   ALLaM powers Saudi Arabia's popular HUMAIN Chat conversational assistant and emphasizes Arabic fluency and cultural nuance. Its series of models is designed to support the ecosystem of Arabic Language Technologies (ALT).

*   Falcon, Abu Dhabi's flagship open LLM series, ranges from 7B to 180B parameters and combines commercially permissive licensing with strong performance that prioritizes resource efficiency – making it popular for academic, commercial, and government uses.

### Asia-Pacific: Scale, Localization, and Linguistic Diversity

The AI ecosystem in the Asia-Pacific region spans the largest and most linguistically diverse populations in the world. Both government-backed initiatives and major tech firms are creating models that serve dozens of languages and dialects while delivering performance that in many cases stands on par with the most advanced models worldwide.

*   [DeepSeek](https://vast.ai/article/serving-deepseek-models-on-vast-ai-with-vllm-and-langchain), one of the most capable open-source models to emerge from China, uses a Mixture-of-Experts (MoE) architecture to balance accuracy and computational efficiency. It offers more fluent support for Asian languages, with cultural nuance that may be missed by Western-centric models.

*   [Qwen](https://vast.ai/article/alibabas-qwen-an-open-source-ai-model-that-surpasses-deepseek), developed by Alibaba, covers 119 languages and dialects, with specialized multimodal variants and built-in support for Model Context Protocol. It features hybrid "thinking" and "non-thinking" modes for flexible reasoning – and you can even set a "thinking budget" to control how much computational effort it devotes to reasoning versus how quickly it replies.

*   tzusumi is a lightweight Japanese LLM (0.6B and 7B parameters) designed for on-prem use, offering strong Japanese-language processing and multimodal support with minimal hardware requirements.

*   Solar Pro 2 offers frontier performance in a comparatively compact 31B model, with multilingual capabilities and standout fluency in Korean. It's designed for real-world agent-like workflows across domains.

*   SeaLLM is a multilingual model built for Southeast Asia that outperforms most open-source LLMs across domains like science, physics, and economics in the languages of the region. Notably, it supports non-Latin scripts and low-resource languages like Lao and Khmer.

### Europe: Responsible AI and Data Sovereignty

In Europe, AI development tends to emphasize data sovereignty and compliance. The region has produced a number of LLMs that serve multilingual users and align with the EU AI Act and GDPR while competing with U.S. and Asian counterparts in raw performance.

*   Mistral, developed in France, offers a suite of open and proprietary models for reasoning, multimodal, coding, and other applications, including its flagship Mistral Large 2 and the open-source Devstral family – and even a multilingual chatbot called Le Chat that boasts deep research and image editing capabilities.

*   Luminous by Aleph Alpha is a multilingual model family (13B–70B parameters) built for enterprise and sovereign-AI use. It emphasizes compliance and transparency and provides support for English, German, French, Italian, and Spanish. The company also offers Pharia, a newer generation of models designed for concise, length-controlled responses.

*   BLOOM is a 176 billion-parameter open-access model produced through a year-long volunteer research effort to provide a transparent alternative to proprietary AI models. It can generate text in 46 natural languages and 13 programming languages.

### Other Regions: Emerging Innovation

Outside the major AI centers, new regional efforts are gaining momentum. For instance, Latam-GPT is an open initiative by Chile's nonprofit CENIA to develop a collaborative AI model completely within and for Latin America – trained on regional data and languages. The project's goal is to advance technological independence and cultural representation across the region.

In Africa, InkubaLM (named after the IsiZulu term for a dung beetle) addresses the need for lightweight African-language models. Trained on English and French as well as five African languages – Hausa, Swahili, IsiXhosa, IsiZulu, and Yoruba – the compact 0.4B-parameter model delivers efficient natural-language capabilities without heavy compute demands.

These projects reflect how emerging regional players are similarly building language models that prioritize inclusion and linguistic diversity, broadening the reach of global AI innovation.

### How to Choose a Regional LLM

Choosing a regional model depends on your goals and infrastructure. Here are a few considerations to keep in mind:

*   Look for open-weight models if you're prioritizing sovereignty or on-prem use.

*   Pick the right model size for your infrastructure! Many regional LLMs offer compact variants.

*   Consider dialect coverage and code-switching capabilities (e.g., training on colloquialisms) – along with nuanced cultural awareness – for social and chat applications.

*   If your use case requires meeting strict data protection and compliance standards, EU-built models developed with these priorities in mind may be ideal.

Once you've found the right fit, the next step is putting these models to work!

### The Bottom Line

Regional LLMs reflect how AI is evolving across a global spectrum of languages, cultural norms, regulatory priorities, and specialized technical needs.

Ultimately, they represent a more inclusive approach to model development. In turn, this helps bridge the AI gap for underrepresented groups worldwide. For developers and organizations, these LLMs provide context-aware and culturally fluent performance that's both versatile and cost-efficient.

With platforms like [Vast.ai](https://vast.ai), you don't need massive in-house infrastructure to take full advantage of these models. Our affordable cloud GPU rental makes it easy to train, fine-tune, and deploy regional LLMs at scale – helping you find just the right fit for your use case and bring AI closer to the communities and markets you serve.

> **Start experimenting with regional LLMs on [Vast.ai](https://vast.ai) today!**


From Jais to Qwen, Falcon to BLOOM - a look at how regional large language models are reshaping AI across languages, cultures, and continents.

Global Growth of Regional Large Language Models (LLMs)


Vast.ai GPUs can be seamlessly rented through SkyPilot, an open-source framework that automates AI, LLM, and batch job execution. This integration makes it easy to access Vast.ai's low-cost, high-performance GPUs while SkyPilot handles provisioning, scaling, and cost optimization.

Whether you're training deep learning models, fine-tuning LLMs, or running large-scale batch jobs, this partnership provides an efficient and cost-effective solution for developers everywhere.

## Why This Integration Matters

Running large-scale AI workloads can be expensive and resource-intensive, especially as GPU availability fluctuates across providers. With Vast.ai's extensive GPU supply and SkyPilot's intelligent cloud orchestration, you can enjoy streamlined access to affordable, on-demand compute resources.

SkyPilot automatically selects the most cost-effective and available infrastructure, provisioning GPUs to ensure seamless execution of AI and batch jobs at scale. As the market leader in low-cost cloud GPU rental, Vast.ai provides an ideal foundation for this approach.

## What Is SkyPilot?

Put simply, SkyPilot is a powerful framework for managing cloud workloads in a way that abstracts away infrastructure complexity. It was developed in the UC Berkeley lab that proposed the concept of "sky computing," which aims to make cloud providers function as a single compute pool.

With SkyPilot, you can run AI models and batch jobs efficiently without worrying about infrastructure management. SkyPilot includes improvements to managed jobs and Kubernetes support, faster GPU provisioning, and ready-to-use LLM recipes.

## Vast.ai's Role

You can easily access and manage GPU resources on Vast.ai through SkyPilot. Vast.ai provides a vast supply of cost-effective GPU compute, while SkyPilot automates the selection and provisioning process to match your workload needs – allowing you to run thousands of concurrent jobs and scale efficiently.

With Vast.ai's extensive GPU marketplace, you get reliable, affordable compute on demand. SkyPilot enhances this by simplifying deployment, ensuring a seamless experience while maximizing access and failover protection.

## Getting Started: Practical Tutorial

Now let's walk through a complete example of deploying a language model on Vast.ai using SkyPilot.

### Setting Up the Environment

First, install SkyPilot with Vast.ai support:

```bash
pip install -U "skypilot[vast]"
pip install "vastai-sdk>=0.1.12"
```

Configure your Vast.ai API key:

```bash
# Get your API key from https://vast.ai → Account → API Keys
echo "YOUR_VAST_API_KEY_HERE" > ~/.vast_api_key
chmod 600 ~/.vast_api_key

# Verify setup
sky check
```

Expected output:

```
Vast: enabled ✓
```

### Creating the YAML Configuration

Create a file named `deepseek-r1-inference.yaml`:

```yaml
resources:
  accelerators: H100:1
  cloud: vast
  image_id: docker:vllm/vllm-openai:latest
  disk_size: 100

run: |
  vllm serve deepseek-ai/DeepSeek-R1-0528-Qwen3-8B \
    --host 0.0.0.0 \
    --port 8000 \
    --max-model-len 4096 \
    --gpu-memory-utilization 0.9 \
    --reasoning-parser deepseek_r1
```

This configuration:

- Requests 1x H100 GPU on Vast.ai
- Uses the pre-built vLLM Docker image
- Allocates 100GB disk space for the model
- Serves the DeepSeek R1 model on port 8000

### Deploying the Service

Launch your service with SkyPilot:

```bash
sky launch deepseek-r1-inference.yaml --cluster deepseek-r1
```

This command will:

1. Find an available H100 GPU on Vast.ai
2. Provision the instance
3. Download the DeepSeek R1 model (~15GB)
4. Start the vLLM API server

The deployment takes 5-10 minutes. Monitor progress with:

```bash
sky logs deepseek-r1 --follow
```

Verify deployment:

```bash
sky status deepseek-r1
```

You should see `Status: UP` when ready.

### SSH Port Forwarding

Since direct port mapping isn't available, use SSH tunneling to access your service:

```bash
ssh -L 8000:localhost:8000 deepseek-r1
```

**Keep this terminal open!** The tunnel needs to stay active for API access.

### Testing the API

In a new terminal, test your deployed service:

```bash
# Test health endpoint
curl http://localhost:8000/health

# List available models
curl http://localhost:8000/v1/models
```

#### Python Client Example

```python
import openai

# Configure client for your local tunnel
client = openai.OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="sk-fake-key"  # vLLM doesn't require real API key
)

# Make a simple request
response = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",
    messages=[{
        "role": "user",
        "content": "Write a brief history of Los Angeles in one paragraph."
    }],
    max_tokens=300,
    temperature=0.7
)

print(response.choices[0].message.content)
```

### Managing Your Deployment

```bash
# Check status
sky status deepseek-r1

# Stop cluster (saves money)
sky stop deepseek-r1

# Start stopped cluster
sky start deepseek-r1

# Terminate completely
sky down deepseek-r1
```

## Conclusion

You now have a production-ready AI API running on Vast.ai's cost-effective H100 GPUs, managed through SkyPilot's streamlined interface. This combination provides enterprise-grade AI capabilities at a fraction of traditional cloud costs.

The integration between Vast.ai and SkyPilot represents a powerful solution for developers who need reliable, affordable GPU access without the complexity of manual infrastructure management. Start experimenting with your own AI workloads using this cost-effective stack!


Easily rent Vast.ai’s low-cost GPUs via SkyPilot for seamless AI, LLM, and batch job execution with automated provisioning, scaling, and cost optimization.

Vast.ai GPUs Can Now Be Rented Through SkyPilot


## Introduction

OpenAI has just released the GPT-OSS family of models, marking their return to open-weight model releases. The family includes GPT-OSS-120B (120 billion parameters) and GPT-OSS-20B (20 billion parameters), offering developers access to state-of-the-art language capabilities previously locked behind proprietary APIs.

GPT-OSS offers configurable reasoning capabilities with reasoning levels - from "low" for quick responses to "high" for complex problem-solving. The model uses OpenAI's harmony encoding system, which structures conversations and reasoning in a sophisticated format that enables fine-grained control over the model's thinking process.

This breakthrough means you can now:

-   Deploy OpenAI-quality models on your own infrastructure
-   Scale inference according to your exact needs

Vast.ai provides the perfect platform for running GPT-OSS models. With its marketplace of GPUs, you can choose the right hardware for your needs - whether running the efficient 20B model or the powerful 120B variant.

In this guide, we'll show you how to deploy both GPT-OSS models on Vast.ai using vLLM for optimized inference, with a focus on the 120B model. You'll learn how to interact with these models using the harmony encoding system for different reasoning levels.

## Setting Up the Environment

Before we can deploy our model, we need to set up our Vast.ai environment. First, install the Vast SDK:

```bash
# Install required packages
pip install --upgrade vastai
pip install --upgrade openai
pip install --upgrade openai-harmony
```

Set up your Vast API key (available from your [Account Page](https://cloud.vast.ai/account/)):

```bash
# Set your Vast.ai API key
export VAST_API_KEY="" # Your key here
vastai set api-key $VAST_API_KEY
```

## Choosing the Right Hardware

The GPT-OSS models have different hardware requirements:

-   **GPT-OSS-120B**: Requires an H100 GPU with 80GB VRAM
-   **GPT-OSS-20B**: Requires 16GB+ VRAM

For this guide, we'll demonstrate with the 120B model on H100. Let's search for suitable instances:

```bash
# Search for suitable GPU instances
vastai search offers " \
gpu_name = H100_SXM \
geolocation=US \
static_ip = true \
direct_port_count >= 1 \
verified = true \
disk_space >= 120 \
rentable = true"
```

## Deploying GPT-OSS with vLLM

Now let's deploy our GPT-OSS instance using vLLM's optimized inference server. We'll use the latest vLLM image with special GPT-OSS support. The pip install commands are from the model's Hugging Face pages ([120B](https://huggingface.co/openai/gpt-oss-120b), [20B](https://huggingface.co/openai/gpt-oss-20b)).

### Deploying GPT-OSS-120B (H100 recommended)

```bash
# Deploy vLLM instance for 120B model
export INSTANCE_ID= # Insert instance ID from search results

vastai create instance $INSTANCE_ID --image vllm/vllm-openai:latest --env '-p 8000:8000' --disk 160  --onstart-cmd 'uv pip install --system --upgrade transformers kernels torch openai; uv pip install --system --pre vllm==0.10.1+gptoss --extra-index-url https://wheels.vllm.ai/gpt-oss/ --extra-index-url https://download.pytorch.org/whl/nightly/cu128; vllm serve openai/gpt-oss-120b --max-model-len 80000'
```

### Alternative: Deploying GPT-OSS-20B on A100

Here's how to deploy the 20B model on an A100 GPU:

```bash
# Deploy vLLM instance for 20B model
export INSTANCE_ID= # Insert instance ID from search results

vastai create instance $INSTANCE_ID --image vllm/vllm-openai:latest --env '-p 8000:8000' --disk 80  --onstart-cmd 'uv pip install --system --upgrade transformers kernels torch openai; uv pip install --system --pre vllm==0.10.1+gptoss --extra-index-url https://wheels.vllm.ai/gpt-oss/ --extra-index-url https://download.pytorch.org/whl/nightly/cu128; vllm serve openai/gpt-oss-20b'
```

After deployment, wait for the model to download and start serving. You can monitor the logs to see when it's ready. Once running, find your instance's IP address and port from the [Instances tab](https://cloud.vast.ai/instances/) in the Vast AI Console.

## Interacting with GPT-OSS Using Harmony Encoding

GPT-OSS uses OpenAI's harmony encoding system, which provides structured conversation formatting and enables different reasoning levels. Let's set up our client:

```python
from openai import OpenAI
from openai_harmony import (
    load_harmony_encoding,
    HarmonyEncodingName,
    Role,
    Message,
    Conversation,
    SystemContent,
    DeveloperContent,
)

# Your server details
VAST_IP_ADDRESS = ""
VAST_PORT = ""

# Initialize client
client = OpenAI(
    api_key="EMPTY",
    base_url=f"http://{VAST_IP_ADDRESS}:{VAST_PORT}/v1"
)

# Load harmony encoding
enc = load_harmony_encoding(HarmonyEncodingName.HARMONY_GPT_OSS)

## Simple Chat Function
def chat_gpt_oss(prompt, reasoning="medium"):
    """
    Simple chat with GPT-OSS
    reasoning: "low", "medium", or "high"
    """
    # Create conversation with harmony format
    conversation = Conversation.from_messages([
        Message.from_role_and_content(Role.SYSTEM, SystemContent.new()),
        Message.from_role_and_content(
            Role.DEVELOPER,
            DeveloperContent.new().with_instructions(f"Reasoning: {reasoning}")
        ),
        Message.from_role_and_content(Role.USER, prompt),
    ])

    # Get formatted prompt
    formatted_prompt = enc.render_conversation_for_completion(conversation, Role.ASSISTANT)

    # Make API call
    response = client.completions.create(
        model="openai/gpt-oss-120b",
        prompt=formatted_prompt,
        max_tokens=2000,
        temperature=0.7,
    )

    return response.choices[0].text


prompt = "Explain quantum computing to a 4th grader. Keep in concise."
```

## Comparing Reasoning Levels

One of GPT-OSS's most interesting features is its configurable reasoning levels. Let's see how the model responds differently to the same prompt with low and high reasoning:

### Low Reasoning Response

```python
result = chat_gpt_oss(prompt, "low")
print("----LOW REASONING----")
print(result)
```

Output:

```
----LOW REASONING----
analysisWe need to explain quantum computing to a 4th grader, concise. Use simple analogies, like bits as light switches, qubits as coins that can be heads, tails, or both. Keep short.assistantfinal**Quantum Computing in a Nutshell (for a 4th‑grader)**

1. **Regular computers = light switches**
   - A normal computer works with bits that are either **0** (off) or **1** (on), just like a light switch that's either down or up.

2. **Quantum computers = magic coins**
   - A quantum computer uses **qubits**.
   - Think of a qubit like a coin that can be **heads (0)**, **tails (1)**, **or both at the same time** while it's spinning in the air. That "both at once" is called **superposition**.

3. **Working together**
   - When you have many regular bits, you have to try every possible combination one after another.
   - With many qubits, they can be in many combinations **all at once** because they can be linked together in a special way called **entanglement**. It's like a group of magic coins that always know what the others are doing, even if they're far apart.

4. **Why it matters**
   - Because of superposition and entanglement, a quantum computer can solve some puzzles (like cracking codes or simulating molecules) **much faster** than a regular computer.

**Bottom line:** A regular computer is like a row of on/off switches; a quantum computer is like a bunch of spinning magic coins that can be many things at once, letting it try many solutions at the same time.
```

### High Reasoning Response

```python
result = chat_gpt_oss(prompt, "high")
print("----HIGH REASONING----")
print(result)
```

Output:

```
----HIGH REASONING----
analysisWe need to explain quantum computing to a 4th grader, concise. Use simple analogies, avoid jargon, but keep accurate. Should be short. Provide analogies like coins, dice, superposition, entanglement. Keep concise.assistantfinal**What is a quantum computer?**

Imagine you have a super‑smart robot that can solve puzzles much faster than a regular computer. A quantum computer is that robot, but instead of using normal "on/off" switches (bits) like a regular computer, it uses tiny, magical switches called **qubits**.

### Bits vs. Qubits
- **Bit (regular computer):** Like a tiny light that can be **off (0)** or **on (1)**.
- **Qubit (quantum computer):** Like a tiny spinning top that can be **0**, **1**, **or both at the same time** (this is called *superposition*).

### Why is that cool?
If you have 10 bits, they can show only one of 2¹⁰ = 1,024 possible patterns at a time.
If you have 10 qubits, they can be in all 2¹⁰ = 1,024 patterns **at once**! That lets a quantum computer try many solutions together.

### A Simple Analogy
Think of a maze puzzle:

- **Regular computer:** Sends a single mouse down the maze, step by step, until it finds the exit.
- **Quantum computer:** Sends a whole swarm of mice that can explore every possible path at the same time and tell you the exit almost instantly.

### Entanglement (Teamwork)
Sometimes two qubits become "best friends" and link together, no matter how far apart they are. Changing one instantly changes the other. This is called **entanglement**, and it lets quantum computers coordinate their work in a super‑efficient way.

### Bottom Line
A quantum computer uses qubits that can be 0, 1, or both at once, and they can be entangled with each other. This lets the machine look at many possibilities at the same time, solving certain problems way faster than ordinary computers.

So, it's like a super‑fast, super‑clever puzzle‑solver that can try lots of answers all at once!
```

## Comparing Low vs High Reasoning Outputs

The quantum computing explanations demonstrate the distinct capabilities of GPT-OSS-120B's reasoning levels. With low reasoning, the model produced a concise, structured explanation using simple analogies like "magic coins" and "light switches" - delivering clear concepts in a compact format suitable for quick understanding.

The high reasoning response revealed significantly more sophisticated thinking. The model provided detailed analogies like the maze puzzle with mice, explained mathematical concepts (2¹⁰ patterns), and structured the explanation with clear sections and formatting. Most notably, it offered multiple perspectives on the same concepts - from the spinning top analogy to the teamwork explanation of entanglement.

This difference in reasoning depth represents a significant advancement in model control. Rather than simply adjusting temperature or other parameters, GPT-OSS allows direct specification of how thoroughly the model should think through problems. The low reasoning approach prioritized efficiency while maintaining accuracy, while high reasoning prioritized comprehensive understanding with rich examples and detailed explanations.

## Conclusion

OpenAI's first open-source model thus introduces not just accessibility to advanced language capabilities, but a new paradigm for controlling model reasoning depth - enabling applications to dynamically adjust between quick responses and thorough analysis based on context and requirements.


Learn how to deploy and run OpenAI's GPT-OSS models on Vast.ai using vLLM, with configurable reasoning levels and harmony encoding system.

Running OpenAI's GPT-OSS on Vast.ai


The DeepSeek R1-0528 release introduces a significant improvement to the reasoning capabilities of language models by making it easier to access "thinking" mode without requiring complex prompt engineering or pre-pending thinking tokens. This enhanced model provides transparent, step-by-step reasoning that's particularly valuable for educational applications, complex problem solving, and scenarios where transparency in AI decision-making is crucial.

In this post, we'll explore how to deploy the **DeepSeek-R1-0528-Qwen3-8B** model using **vLLM** on Vast.ai's cloud GPU platform, leveraging the new **qwen3 reasoning parser** that simplifies access to the model's internal thinking process.

---

## The Open Source AI Revolution

DeepSeek R1-0528 represents more than just a technical advancement—it's a significant step toward democratizing advanced AI capabilities. As an **open source model**, it makes these reasoning and advanced AI capabilities more accessible.

Unlike closed-source models that require vendor dependence, usage costs, and data privacy concerns, DeepSeek R1-0528 enables **complete control** over your AI infrastructure. Organizations can now deploy advanced reasoning capabilities locally, maintaining data sovereignty while avoiding variable per-token pricing.

This democratization means organizations can access cutting-edge AI without massive infrastructure investments. Advanced reasoning capabilities are no longer exclusive to large tech companies—anyone with GPU access can deploy their own reasoning-capable AI system.

---

## What Makes DeepSeek R1-0528 Special?

The DeepSeek R1-0528 release addresses a key limitation in previous DeepSeek models: the complexity of accessing the model's internal reasoning process. Previously, users needed to manually prepend "thinking" tokens to model outputs or use complex prompt engineering to see step-by-step reasoning.

---

## Deploying DeepSeek R1-0528 on Vast.ai

### Step 1: Set Up Vast.ai Environment

First, install the Vast CLI and configure your API key:

```bash
pip install --upgrade vastai

export VAST_API_KEY="your_vast_api_key"
vastai set api-key $VAST_API_KEY
```

### Step 2: Find Suitable Hardware

The DeepSeek-R1-0528-Qwen3-8B model requires specific hardware capabilities:

-   **Minimum 24GB GPU RAM** for model weights and KV cache
-   **Single GPU configuration** (sufficient for the 8B parameter model)
-   **Static IP address** for stable API endpoint hosting
-   **Direct port access** for the vLLM OpenAI-compatible server
-   **60GB+ disk space** for model storage and dependencies

Search for suitable instances:

```bash
vastai search offers "compute_cap >= 750 \
gpu_ram >= 24 \
num_gpus = 1 \
static_ip = true \
direct_port_count >= 1 \
verified = true \
disk_space >= 60 \
rentable = true"
```

### Step 3: Deploy the Model

Deploy using the vLLM OpenAI-compatible server with the qwen3 reasoning parser:

```bash
export INSTANCE_ID="your_instance_id"

vastai create instance $INSTANCE_ID \
--image vllm/vllm-openai:latest \
--env '-p 8000:8000' \
--disk 60 \
--args --model deepseek-ai/DeepSeek-R1-0528-Qwen3-8B \
--max-model-len 4096 \
--reasoning-parser qwen3
```

**Key deployment parameters:**

-   `--reasoning-parser qwen3`: Enables the simplified reasoning parser
-   `--max-model-len 4096`: Accommodates longer reasoning sequences
-   `--tensor-parallel-size 1`: Single GPU configuration
-   `--gpu-memory-utilization 0.90`: Optimized memory usage

### Step 4: Get Instance Details and Test Connection

Navigate to the [Instances tab](https://cloud.vast.ai/instances/) in the Vast AI Console and locate your instance. Click the IP address button to view the forwarded ports.

Test the connection with a simple curl command:

```bash
export VAST_IP_ADDRESS="your_ip_address"
export VAST_PORT="your_port"

curl -X POST http://$VAST_IP_ADDRESS:$VAST_PORT/v1/completions \
-H "Content-Type: application/json" \
-d '{"model": "deepseek-ai/DeepSeek-R1-0528-Qwen3-8B", "prompt": "Hello, how are you?", "max_tokens": 50}'
```

---

## Testing the Model's Reasoning Capabilities

### Setup Python Client

```python
import time
from openai import OpenAI

# Replace with your actual Vast.ai instance details
VAST_IP_ADDRESS = "your_ip_address"
VAST_PORT = "your_port"

client = OpenAI(
    api_key="DUMMY",
    base_url=f"http://{VAST_IP_ADDRESS}:{VAST_PORT}/v1"
)
```

### Basic Functionality Test

```python
start = time.time()
test_resp = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",
    messages=[{"role": "user", "content": "Hello"}]
)
elapsed = time.time() - start

print("Response:", test_resp.choices[0].message.content)
print(f"Elapsed time: {elapsed:.4f} s")
```

### Reasoning Capability Test

This is where DeepSeek R1-0528 truly shines. The model naturally provides step-by-step reasoning without complex prompt engineering:

```python
reasoning_prompt = """
Write the History of Paris in a paragraph, show your thinking
step by step first before paragraph and then write the history
in a paragraph
"""

start = time.time()
resp = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",
    messages=[{"role": "user", "content": reasoning_prompt}]
)
elapsed = time.time() - start

print("Model output:\n", resp.choices[0].message.content)
print(f"Elapsed time: {elapsed:.3f} s")
```

**Expected output structure:**

1. **Thinking process**: The model shows its step-by-step reasoning
2. **Final answer**: A well-structured paragraph about Paris's history

### Mathematical Reasoning Test

```python
math_prompt = "If 8 × 7 = ?, show your work step by step."

resp = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",
    messages=[{"role": "user", "content": math_prompt}]
)

print("Mathematical reasoning:\n", resp.choices[0].message.content)
```

The model will naturally break down the problem, show intermediate steps, and arrive at the solution with clear reasoning traces.

---

## Key Benefits of DeepSeek R1-0528

### 1. **Simplified Reasoning Access**

No need to prepend thinking tokens or use complex prompt engineering. The `qwen3` reasoning parser automatically handles the extraction and presentation of reasoning steps.

### 2. **Educational Value**

Perfect for educational applications where students need to understand the problem-solving process, not just the final answer.

### 3. **Transparency and Trust**

Users can see exactly how the model arrives at its conclusions, building trust in AI-powered decision-making systems.

### 4. **Cost-Effective Deployment**

Running on Vast.ai provides significant cost savings compared to major cloud providers while maintaining high performance.

### 5. **Easy Integration**

OpenAI-compatible API ensures seamless integration with existing applications and workflows.

---

## Use Cases

### Educational Applications

-   Step-by-step problem solving for mathematics, science, and logic
-   Teaching critical thinking and reasoning skills
-   Creating interactive learning experiences

### Business Intelligence

-   Transparent decision-making in automated systems
-   Audit trails for AI-powered recommendations
-   Complex analysis with clear reasoning paths

### Research and Development

-   Understanding model behavior and capabilities
-   Testing reasoning patterns across different domains
-   Developing more interpretable AI systems

---

## Conclusion

The DeepSeek R1-0528 release represents a significant step forward in making AI reasoning more accessible and transparent. By combining this powerful model with Vast.ai's cost-effective GPU infrastructure and the simplified qwen3 reasoning parser, developers can now easily deploy transparent, reasoning-capable AI systems.

Whether you're building educational tools, business intelligence systems, or research applications, DeepSeek R1-0528 on Vast.ai provides an excellent foundation for applications that require clear, step-by-step reasoning capabilities.

The elimination of complex prompt engineering for accessing reasoning modes makes this model particularly valuable for production applications where transparency and interpretability are key requirements.

---

## Useful Resources

-   [DeepSeek R1-0528 Model on Hugging Face](https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B)
-   [vLLM Documentation](https://docs.vllm.ai/)
-   [Vast.ai GPU Marketplace](https://vast.ai)
-   [OpenAI Python SDK](https://github.com/openai/openai-python)

---

_Ready to deploy transparent, reasoning-capable AI? Get started with DeepSeek R1-0528 on Vast.ai today! 🚀_


Deploy DeepSeek R1-0528 on Vast.ai to access transparent, step-by-step AI reasoning without prompt tricks—ideal for education, research, and business use.

Posts about: Vast.ai

Vast.ai Named Among Fastest Growing Vendors by Ramp and Brex

January & February 2026 Product Updates

Global Growth of Regional Large Language Models (LLMs)

Vast.ai GPUs Can Now Be Rented Through SkyPilot

Running OpenAI's GPT-OSS on Vast.ai

DeepSeek R1-0528: Enhanced Reasoning with Simplified Thinking Mode on Vast.ai