We run a multi-model AI API with 399 models. Every day we compare our prices against OpenRouter for the same models, same tokens. Today's snapshot: our basket of 18 flagship models averages 78% below OpenRouter — and on some models the gap is brutal. Here's the full receipt, model by model, with the raw numbers.
Why we publish this
A few days ago, an independent blog post that benchmarked OpenRouter against direct providers hit the front page of Hacker News with 700+ points. The lesson from that post is simple: developers don't trust marketing. They trust measurement.
So instead of telling you we're cheaper, we're going to show you, per model, per 1M tokens, what we charge vs what the same model costs at OpenRouter (the OpenRouter list price). This data is generated automatically every day from our pricing engine — nothing here is hand-picked.
The headline numbers (September 12, 2026)
Our daily Price Index tracks a fixed basket of 18 flagship models. Today:
- Average discount vs OpenRouter: 78.05%
- Best deal in the basket: DeepSeek V4 Pro at 98.2% off
- 15 of 18 models: 50%+ below OpenRouter
The full receipt, model by model
All prices are per 1M tokens (input/output):
| Model | Qubax in/out | OpenRouter in/out | You save |
|---|---|---|---|
| DeepSeek V4 Pro | $0.0097 / $0.039 | $0.87 / $1.20 | 98.2% |
| Qwen3 235B A22B | $0.0025 / $0.011 | $0.09 / $0.35 | 97.0% |
| Kimi K2.5 | $0.018 / $0.087 | $0.45 / $2.25 | 96.1% |
| GPT-4o | $0.11 / $0.44 | $2.50 / $10.00 | 95.6% |
| GLM 5.2 | $0.020 / $0.079 | $0.49 / $1.56 | 95.5% |
| Gemini 3 Flash | $0.015 / $0.09 | $0.25 / $1.50 | 94.0% |
| DeepSeek V4 Flash | $0.0034 / $0.014 | $0.068 / $0.168 | 93.6% |
| GPT-5.3 Codex | $0.13 / $1.05 | $1.75 / $14.00 | 92.5% |
| Llama 3.3 70B | $0.009 / $0.029 | $0.10 / $0.32 | 91.0% |
| Grok 4.20 Beta | $0.19 / $0.38 | $1.25 / $2.50 | 85.0% |
| Gemini 3.1 Pro | $0.15 / $0.90 | $1.00 / $6.00 | 85.0% |
| GPT-5.2 | $0.13 / $1.05 | $0.88 / $7.00 | 85.0% |
| GPT-5.4 Mini | $0.068 / $0.41 | $0.38 / $2.25 | 82.0% |
| GPT-5.4 | $0.23 / $1.35 | $1.25 / $7.50 | 82.0% |
| DeepSeek R1 | $0.20 / $0.71 | $0.70 / $2.50 | 71.5% |
| Claude Opus 4.8 | $2.20 / $11.02 | $5.00 / $25.00 | 55.9% |
| Claude Sonnet 4.5 | $1.35 / $6.75 | $3.00 / $15.00 | 55.0% |
Honesty corner: the one model where we cost more
Grok 4.1 Fast currently prices above OpenRouter on our marketplace (−50%). This happens when our upstream liquidity is thin for a model. We don't hide it — if a model is cheaper elsewhere, use it there. Our job is to be the cheapest option on most models most of the time, and to tell you when we're not.
What this means for your bill
Concrete example: an agentic coding workload burning 50M input + 10M output tokens per week:
- At OpenRouter (GPT-5.4): ~$137.50/week → ~$590/month
- On Qubax (GPT-5.4): ~$24.75/week → ~$106/month
Same model, same tokens, same API shape (OpenAI-compatible). The difference is purely where you buy.
How to verify us
Don't take our word for it. Our live pricing page shows per-model prices with the OpenRouter comparison, and the AI Model Price Index updates daily. Run your own benchmark: same prompt, same model, both APIs, compare the token-level costs from the response headers. We win on receipts, not adjectives.
Qubax gives you 399 models behind one OpenAI-compatible API — no subscription, pay-as-you-go with crypto or card, no KYC. Create a free API key and run the comparison yourself.
