Pricing · Arbitrage7 models · 34 provider listings · Up to 10x price spread
AI Arbitrage · Same Model, 10x Price Spread
Every open-weight model hosted on multiple providers. Same weights, different prices. Pick the cheapest with zero quality trade-off.
Llama 4
405B · Llama Community License
Meta's flagship open-weight model · runs on six major inference providers with 10x price spread.
Cheapest
Groq$0.030/M
Save vs most expensive
86%
$0.22 → $0.030
D
See all prices
Qwen3.5 397B-A17B
397B (17B active) · Apache 2.0
Alibaba's flagship MoE model · 397B total / 17B active · multilingual leader.
Cheapest
Alibaba Cloud$0.280/M
Save vs most expensive
53%
$0.60 → $0.280
D
H
See all prices
DeepSeek V3.2
671B (37B active) · MIT
MoE reasoning model · rewrites cost structure of frontier-tier inference.
Cheapest
DeepSeek (direct)$0.140/M
Save vs most expensive
81%
$0.75 → $0.140
H
See all prices
gpt-oss-20b
20B · Apache 2.0
OpenAI's open-weight release · Apache 2.0 · self-hostable on consumer GPUs.
Cheapest
Cerebras$0.040/M
Save vs most expensive
67%
$0.12 → $0.040
D
See all prices
gpt-oss-120b
120B · Apache 2.0
OpenAI's larger open release · reasoning-capable · still cheap on optimized stacks.
Cheapest
Groq$0.150/M
Save vs most expensive
63%
$0.40 → $0.150
D
See all prices
Mistral Large 3
123B · Mistral Research License
Mistral's flagship · strong in code + multilingual · EU-hosted primary.
Cheapest
Groq$1.80/M
Save vs most expensive
28%
$2.50 → $1.80
AA
See all prices
Llama 4 Maverick
170B (17B active) · Llama Community License
Meta's multimodal Llama 4 variant · vision + text · long context.
Cheapest
Fireworks AI$0.180/M
Save vs most expensive
55%
$0.40 → $0.180
D
See all prices
Frequently Asked Questions
When the same open-weight model (Llama 4, DeepSeek V3.2, Qwen3.5, etc.) is hosted on multiple providers at different prices. Because the weights are identical, you can pick the cheapest with zero quality trade-off. Arbitrage spreads of 5-10x are common.