GPT OSS 120BvsMercury
GPT OSS 120B is cheaper than Mercury at $0.18/M vs $10/M input tokens.
| Spec | GPT OSS 120B | Mercury |
|---|---|---|
| Input price | $0.18/M tokens | $10/M tokens |
| Output price | $0.8/M tokens | $10/M tokens |
| Context window | 131K tokens | 32K tokens |
| Parameters | 117B (5.1B active) | Not disclosed |
| Weights | Open | — |
| Free API (OpenRouter) | No | No |
| Released | Aug 2025 | Jun 2025 |
Our Verdict
GPT OSS 120B
MercuryRunner-up
Not enough votes to call it. On the specs, GPT OSS 120B has the edge: bigger model tier, newer, bigger context window, major provider backing.
GPT OSS 120B wins Web Design and Image Generation. GPT OSS 120B costs 13x less per token.
Slight edge
API pricing
Cost per 1M tokens
GPT OSS 120B
Input
$0.18
56× cheaper
Output
$0.80
13× cheaper
Mercury
Input
$10.00
Output
$10.00
GPT OSS 120B is cheaper on both: 56× input, 13× output.
Where to run it
18 hosts, cheapest first
GPT OSS 120B18 hosts
AkashMLbf16$0.03 in$0.17 out131k99.9% upCoreWeavefp4$0.03 in$0.17 out131k99.9% upDeepInfrabf16$0.04 in$0.17 out131k99.4% upDigitalOcean$0.06 in$0.39 out128k99.7% upMancerfp8$0.06 in$0.50 out131k99.6% upGoogle Vertex AI$0.09 in$0.36 out131k99.9% up
12 more hostsFewer hosts
Basetenfp4$0.10 in$0.50 out128k100% upParasailfp4$0.10 in$0.75 out131k100% upSambaNova$0.14 in$0.95 out131k96.9% upAmazon Bedrock$0.15 in$0.60 out131k98.1% upGroq$0.15 in$0.60 out131k100% upNebiusfp4$0.15 in$0.60 out131k96.2% upTogether$0.15 in$0.60 out131k98.8% upCerebrasfp16$0.35 in$0.75 out131k100% upNovitafp4degraded$0.05 in$0.25 out131k86.7% upSiliconFlowfp8degraded$0.05 in$0.45 out131k84.3% upPhaladegraded$0.15 in$0.60 out131k88.8% upMaradegraded$0.15 in$0.75 out131k86% up
Mercury
No hosts listed on OpenRouter.
Per million tokens. Prices and uptime via OpenRouter, checked 9 Sep 2026.
Writing DNA
Style Comparison
Similarity
62%
GPT OSS 120B uses 5.5x more emoji
GPT OSS 120B
Mercury
52%Vocabulary60%
19wSentence Length31w
0.28Hedging0.78
7.4Bold4.1
1.8Lists2.8
0.15Emoji0.03
0.73Headings0.88
0.17Transitions0.06
Based on 21 + 26 text responses
Research
What we learned reading every model
FAQ
Common questions
Keep exploring
