GLM 4.6vsGPT OSS 120B
GLM 4.6 wins 76% of 46 blind community votes against GPT OSS 120B on Rival.
| Spec | GLM 4.6 | GPT OSS 120B |
|---|---|---|
| Win rate | 76% | 24% |
| Input price | $0.4/M tokens | $0.18/M tokens |
| Output price | $1.75/M tokens | $0.8/M tokens |
| Context window | 203K tokens | 131K tokens |
| Weights | Open | Open |
| Free API (OpenRouter) | No | No |
| Released | Sep 2025 | Aug 2025 |
Our Verdict
GLM 4.6Winner
GPT OSS 120BRunner-up
Pick GLM 4.6. In 46 blind votes, GLM 4.6 wins 76% of the time. That's not luck.
GLM 4.6 wins 3 categories, Image Generation by the widest margin.
Clear winner
Reviewing agent-written code?See a Brief PR report
API pricing
Cost per 1M tokens
GLM 4.6
Input
$0.40
Output
$1.75
GPT OSS 120B
Input
$0.18
2.2× cheaper
Output
$0.80
2.2× cheaper
GPT OSS 120B is cheaper on both: 2.2× input, 2.2× output.
Where to run it
24 hosts, cheapest first
GPT OSS 120B19 hosts
AkashMLbf16$0.03 in$0.17 out131k98.8% upCoreWeavefp4$0.03 in$0.17 out131k99.9% upDekaLLMbf16$0.03 in$0.18 out131k99.2% upDeepInfrabf16$0.04 in$0.17 out131k98.7% upDigitalOcean$0.06 in$0.39 out128k99.3% upMancerfp8$0.06 in$0.50 out131k99% up
13 more hostsFewer hosts
Google Vertex AI$0.09 in$0.36 out131k99.9% upBasetenfp4$0.10 in$0.50 out128k100% upParasailfp4$0.10 in$0.75 out131k97.8% upSambaNova$0.14 in$0.95 out131k98% upAmazon Bedrock$0.15 in$0.60 out131k99.2% upGroq$0.15 in$0.60 out131k100% upNebiusfp4$0.15 in$0.60 out131k98.5% upSiliconFlowfp8$0.15 in$0.60 out131k95.4% upTogether$0.15 in$0.60 out131k96.5% upCerebrasfp16$0.35 in$0.75 out131k100% upNovitafp4degraded$0.05 in$0.25 out131k96% upPhaladegraded$0.15 in$0.60 out131k96.6% upMaradegraded$0.15 in$0.75 out131k60.9% up
Per million tokens. Prices and uptime via OpenRouter, checked 11 Sep 2026.
Writing DNA
Style Comparison
Similarity
61%
GPT OSS 120B uses 2.1x more transitions
GLM 4.6
GPT OSS 120B
52%Vocabulary52%
16wSentence Length19w
0.31Hedging0.28
5.1Bold7.4
3.2Lists1.8
0.14Emoji0.15
0.54Headings0.73
0.08Transitions0.17
Based on 28 + 21 text responses
Research
What we learned reading every model
FAQ
