Together AI
$4.50 / month- Input cost
- $1.50
- Output cost
- $3.00
- Per 1,000 calls
- $0.45
- Cache pricing
- Not published
Same model, different API route
GPT-OSS 120B: Together AI or OpenRouter for lower API cost?
Compare Together AI's fixed serverless endpoint with OpenRouter's current catalog floor for GPT-OSS 120B. This comparison holds the model and token workload constant; it does not assume identical infrastructure, latency, or service limits.
Official route fields
Provider pricing and route limits
Prices are USD per 1M tokens. “Not published” means the official tracked source does not expose that field.
| Field | Together AI | OpenRouter |
|---|---|---|
| Pricing basis | Fixed provider endpoint | Current catalog floor |
| API model ID | openai/gpt-oss-120b | openai/gpt-oss-120b |
| Input / 1M | $0.15 | $0.037 |
| Cached input / 1M | Not published | Not published |
| Output / 1M | $0.60 | $0.17 |
| Published route context | 128,000 tokens | 131,072 tokens |
| Published maximum output | Not published | Not published |
Editable workload
GPT-OSS 120B monthly API cost
Token inference only. Provider payment, BYOK, tool, storage, and untracked request fees are excluded.
OpenRouter
$1.22 / month- Input cost
- $0.37
- Output cost
- $0.85
- Per 1,000 calls
- $0.122
- Cache pricing
- Not published
OpenRouter is $3.28 lower per month for the default workload. This is a route-price result, not a service-quality ranking.
Additional fee boundary: OpenRouter credit purchases currently carry a separate 5.5% fee with a minimum charge. The token estimate below does not add payment or BYOK fees.
What to test before switching routes
- The same API model string does not guarantee identical infrastructure or service limits.
- OpenRouter credit-purchase and BYOK fees remain outside the token-only estimate.
- Measure error handling, latency, and rate-limit behavior with representative traffic.
What this comparison controls
Both routes call OpenAI's GPT-OSS 120B using the published API model string. Provider infrastructure, quantization, request limits, routing, retention controls, and operational behavior can still differ.
Related provider routes
Continue comparing GPT-OSS 120B
Keep the model constant and compare the remaining provider pair.