DeepInfra API Pricing & Inference Rates
DeepInfra delivers rock-bottom token pricing on open-source AI models, including Meta Llama 3.3, DeepSeek-V3, Qwen 2.5, and Whisper. Explore model rate cards, serverless GPU pricing, and unit economics.
DeepInfra Monthly Compute Estimator
Calculate monthly bills across DeepInfra serverless endpoints for open-weight LLMs.
Production Pricing & Model Tier Rates
Verified rates, context capacities, typical inference latency, and recommended enterprise workloads.
| Model / Tier | Input Rate (1M) | Output Rate (1M) | Context Window | Latency / Throughput | Recommended Workload |
|---|---|---|---|---|---|
| Llama 3.3 70B Turbo | $0.230 | $0.400 | 128,000 | 60–90 tok/s | High-volume reasoning, summarization, synthetic data |
| Llama 3.1 8B Instant | $0.055 | $0.055 | 128,000 | 140–180 tok/s | Classification, lightweight agent routing, JSON extraction |
| DeepSeek-V3 (671B MoE) | $0.550 | $1.100 | 64,000 | 40–70 tok/s | Full-codebase refactoring, agentic programming |
| DeepSeek-R1 (Reasoning) | $0.800 | $2.400 | 64,000 | 30–50 tok/s | Complex math, competitive coding, formal logic |
| Qwen 2.5 Coder 32B | $0.200 | $0.200 | 32,768 | 80–110 tok/s | High-speed code completion, SWE-bench tasks |
| Whisper Large v3 (Audio) | $0.0015 / min | ($0.090 / hr) | N/A | Fast batch | Speech-to-text batch transcription at 75% off OpenAI |
⚡ Architectural Deep Dive & Unit Economics Analysis
DeepInfra has emerged as the premier value arbitrage play in modern AI infrastructure. By managing bare-metal GPU clusters across geographically distributed tier-3 and tier-4 data centers running optimized vLLM and TensorRT-LLM runtimes, DeepInfra slashes inference margins. For instance, hosting Llama 3.3 70B costs only $0.23 per million input tokens and $0.40 per million output tokens—making it approximately 55% cheaper than Together AI and Fireworks AI, and 75% cheaper than AWS Bedrock.
Beyond standard chat completions, DeepInfra provides serverless endpoints for open embeddings (BGE, E5), text-to-speech, and automatic speech recognition (Whisper Large v3 at $0.09/hour). For engineering teams building high-throughput data extraction or synthetic training dataset generation pipelines, routing traffic to DeepInfra delivers dramatic budget relief.
Production Workloads & Monthly Cost Modeling
Evaluating practical unit economics across real-world application architectures.
Continuous Synthetic Dataset Generation
Generating 200M prompt tokens and 80M output tokens monthly for model distillation. DeepInfra runs at $78.00/mo vs Together AI's $246.40/mo.
High-Volume Podcast Speech Transcription
Transcribing 1,000 hours of monthly user audio with Whisper Large v3. DeepInfra charges $90.00 flat versus OpenAI's $360.00, saving 75%.
Automated Web Scraping Entity Extraction
Extracting structured JSON from scraped web pages using Llama 3.1 8B. 100M tokens cost $5.50 on DeepInfra versus $15.00 on GPT-4o-mini.
import os
from openai import OpenAI
# DeepInfra is 100% wire-compatible with OpenAI API clients
client = OpenAI(
api_key=os.environ.get("DEEPINFRA_API_KEY"),
base_url="https://api.deepinfra.com/v1/openai"
)
def query_deepinfra_llama(prompt: str):
"""Run Llama 3.3 70B at ultra-low $0.23 / $0.40 per million tokens."""
response = client.chat.completions.create(
model="meta-llama/Llama-3.3-70B-Instruct",
messages=[{"role": "user", "content": prompt}],
temperature=0.2,
max_tokens=2048
)
return response.choices[0].message.contentOperational Gotchas & Failure Modes
Engineering realities, throttling thresholds, and hidden surcharge traps to monitor.
Related Models, Showdowns & Benchmarks
Cross-reference provider rates against hosted models, direct head-to-head battles, and calculators.
Cheapest LLM APIs of 2026
Compare DeepInfra against 40+ low-cost inference providers.
Groq vs Together AI Showdown
Compare LPU throughput speeds against serverless GPU clouds.
Llama 3.3 70B vs DeepSeek-V3
Evaluate Meta open weights against DeepSeek MoE architecture.
LLM Fine-Tuning ROI Calculator
Calculate when fine-tuning your own weights beats API prompts.
Frequently Asked Questions
How does DeepInfra achieve such low pricing?
DeepInfra optimizes GPU utilization by co-locating workloads on efficient bare-metal data centers, leveraging vLLM/TensorRT-LLM PagedAttention kernels, and eliminating high cloud hypervisor markups.
Is DeepInfra compatible with the OpenAI Python SDK?
Yes! DeepInfra provides 100% OpenAI API wire-compatibility. Simply change base_url to 'https://api.deepinfra.com/v1/openai' and provide your DEEPINFRA_API_KEY.
What is DeepInfra's pricing for Llama 3.3 70B?
DeepInfra charges $0.23 per million input tokens and $0.40 per million output tokens for Llama 3.3 70B, making it one of the cheapest hosted 70B endpoints in the world.
Does DeepInfra offer speech-to-text models?
Yes. DeepInfra hosts Whisper Large v3 for $0.0015 per audio minute ($0.090 per hour), which is 75% cheaper than OpenAI's official $0.0060/min Whisper API.
Can I deploy custom models or LoRA adapters on DeepInfra?
Yes. DeepInfra supports deploying custom Hugging Face model repositories and LoRA weights on dedicated GPU instances billed by the second.
Does DeepInfra store or log prompt data?
DeepInfra does not log or store prompt data or completions by default, and does not use customer data to train foundation models.
DeepInfra Serverless Inference Economics: Pay-Per-Token Open-Weight Hosting
DeepInfra specializes in ultra-low-cost, pay-per-token serverless inference for open-weight foundation models (Llama 3.3 70B, Qwen 2.5 Coder 32B, DeepSeek-R1, and Mistral Large). By running optimized FP8 quantization kernels on large multi-GPU clusters, DeepInfra achieves some of the lowest commercial rates in the industry: $0.13 per million input tokens and $0.40 per million output tokens on Llama 3.3 70B. Furthermore, DeepInfra provides serverless Whisper transcription ($0.0005/min) and embedding models ($0.02/M tokens) under a single unified billing balance.
DeepInfra Model Benchmark & Rate Card
| Model Name | Input ($/M) | Output ($/M) | Context Window | Typical Speed |
|---|---|---|---|---|
| meta-llama/Llama-3.3-70B-Instruct | $0.13 | $0.40 | 128,000 tokens | 85 tok/s |
| Qwen/Qwen2.5-Coder-32B-Instruct | $0.07 | $0.16 | 32,000 tokens | 95 tok/s |
| deepseek-ai/DeepSeek-R1 | $0.70 | $2.50 | 64,000 tokens | 65 tok/s |
| openai/whisper-large-v3 | $0.0005 / min | N/A | Batch audio | 15x Real-Time |
DeepInfra Multi-Model Client Implementation
import OpenAI from 'openai';
const client = new OpenAI({
baseURL: 'https://api.deepinfra.com/v1/openai',
apiKey: process.env.DEEPINFRA_TOKEN
});
// Run low-cost code generation tasks at $0.07/M input tokens
export async function generateComponentBoilerplate(prompt: string) {
return await client.chat.completions.create({
model: 'Qwen/Qwen2.5-Coder-32B-Instruct',
messages: [{ role: 'user', content: prompt }]
});
}Engineering Q&A: DeepInfra Production Use
Does DeepInfra charge minimum monthly platform fees?
No. DeepInfra operates strictly on pay-per-token usage with zero monthly seat subscriptions or minimum spending commitments, making it ideal for lean startups and background batch pipelines.
How does DeepInfra compare with Together AI on latency?
Both providers deliver roughly equivalent Time-to-First-Token (150-250ms) on Llama 3.3 70B. DeepInfra is frequently 10% to 20% cheaper on input token pricing, while Together AI provides broader multi-modal fine-tuning tooling.