PRICE~moonshotai/kimi-latest cached_input_per_mtok increased from 0.29 to 0.809m PRICE~moonshotai/kimi-latest output_per_mtok increased from 11.36 to 13.009m PRICE~moonshotai/kimi-latest input_per_mtok increased from 0.75 to 0.989m PRICE~deepseek/deepseek-pro-latest input_per_mtok decreased from 0.13 to 0.1340m PRICE~deepseek/deepseek-pro-latest output_per_mtok increased from 0.40 to 4.2040m PRICE~deepseek/deepseek-pro-latest cached_input_per_mtok increased from 0.0042 to 0.1040m PRICE~moonshotai/kimi-latest input_per_mtok decreased from 0.99 to 0.7540m PRICE~moonshotai/kimi-latest output_per_mtok decreased from 13.00 to 11.3640m PRICE~moonshotai/kimi-latest cached_input_per_mtok decreased from 0.80 to 0.2940m PRICE~moonshotai/kimi-latest input_per_mtok increased from 0.75 to 0.991h PRICE~moonshotai/kimi-latest output_per_mtok increased from 11.36 to 13.001h PRICE~moonshotai/kimi-latest cached_input_per_mtok increased from 0.29 to 0.801h PERFORMANCEGLM-5.2 → Scaleway: TTFT ↑ 20%2h PERFORMANCEDeepSeek V4 Flash → Scaleway: TTFT ↑ 35%2h PERFORMANCEMistral Medium 3.5 → Scaleway: TTFT ↓ 21%2h PERFORMANCEDeepSeek V4 Flash → Scaleway: throughput ↓ 20%2h PERFORMANCEGPT-OSS 120B → Scaleway: throughput ↑ 31%2h PERFORMANCEGPT-OSS 120B → Scaleway: TTFT ↓ 24%2h PERFORMANCEGPT-OSS 120B → Together AI: throughput ↑ 20%2h PERFORMANCEKimi K3 via OpenRouter: TTFT ↓ 49%2h PERFORMANCEGLM-5.2 via OpenRouter: reliability recovered 96.8% → 100.0%2h PERFORMANCEKimi K3 via Cortecs: throughput ↑ 184%2h PERFORMANCEKimi K3 via Cortecs: TTFT ↓ 42%2h PERFORMANCEGLM-5.2 via OpenRouter: throughput ↓ 49%2h PERFORMANCEGLM-5.2 via OpenRouter: TTFT ↑ 121%2h PERFORMANCEMiniMax M3 via OpenRouter: reliability recovered 96.8% → 100.0%2h PERFORMANCEDeepSeek V4 Pro via Cortecs: throughput ↑ 27%2h PERFORMANCEGPT-OSS 120B via OpenRouter: throughput ↓ 22%2h PERFORMANCEGLM-4.7 via OpenRouter: TTFT ↓ 78%2h PERFORMANCEGPT-OSS 20B → Groq: TTFT ↑ 24%2h PERFORMANCEKimi K3 → Together AI: TTFT ↓ 34%2h PERFORMANCEDeepSeek V4 Flash via OpenRouter: throughput ↓ 17%2h PERFORMANCELlama 3.3 70B via Cortecs: TTFT ↓ 51%2h PERFORMANCEQwen3 235B via OpenRouter: throughput ↑ 37%2h PERFORMANCEQwen3 235B via OpenRouter: TTFT ↓ 29%2h PERFORMANCEDeepSeek V4 Flash → Together AI: throughput ↑ 19%2h PERFORMANCEGLM-5.3 via OpenRouter: throughput ↓ 17%2h PERFORMANCEGPT-OSS 20B via Cortecs: throughput ↑ 28%2h PERFORMANCELlama 3.3 70B → Together AI: TTFT ↓ 33%2h PRICEtencent/hy4-preview output_per_mtok increased from 2.25 to 2.502h

PRICE~moonshotai/kimi-latest cached_input_per_mtok increased from 0.29 to 0.809m PRICE~moonshotai/kimi-latest output_per_mtok increased from 11.36 to 13.009m PRICE~moonshotai/kimi-latest input_per_mtok increased from 0.75 to 0.989m PRICE~deepseek/deepseek-pro-latest input_per_mtok decreased from 0.13 to 0.1340m PRICE~deepseek/deepseek-pro-latest output_per_mtok increased from 0.40 to 4.2040m PRICE~deepseek/deepseek-pro-latest cached_input_per_mtok increased from 0.0042 to 0.1040m PRICE~moonshotai/kimi-latest input_per_mtok decreased from 0.99 to 0.7540m PRICE~moonshotai/kimi-latest output_per_mtok decreased from 13.00 to 11.3640m PRICE~moonshotai/kimi-latest cached_input_per_mtok decreased from 0.80 to 0.2940m PRICE~moonshotai/kimi-latest input_per_mtok increased from 0.75 to 0.991h PRICE~moonshotai/kimi-latest output_per_mtok increased from 11.36 to 13.001h PRICE~moonshotai/kimi-latest cached_input_per_mtok increased from 0.29 to 0.801h PERFORMANCEGLM-5.2 → Scaleway: TTFT ↑ 20%2h PERFORMANCEDeepSeek V4 Flash → Scaleway: TTFT ↑ 35%2h PERFORMANCEMistral Medium 3.5 → Scaleway: TTFT ↓ 21%2h PERFORMANCEDeepSeek V4 Flash → Scaleway: throughput ↓ 20%2h PERFORMANCEGPT-OSS 120B → Scaleway: throughput ↑ 31%2h PERFORMANCEGPT-OSS 120B → Scaleway: TTFT ↓ 24%2h PERFORMANCEGPT-OSS 120B → Together AI: throughput ↑ 20%2h PERFORMANCEKimi K3 via OpenRouter: TTFT ↓ 49%2h PERFORMANCEGLM-5.2 via OpenRouter: reliability recovered 96.8% → 100.0%2h PERFORMANCEKimi K3 via Cortecs: throughput ↑ 184%2h PERFORMANCEKimi K3 via Cortecs: TTFT ↓ 42%2h PERFORMANCEGLM-5.2 via OpenRouter: throughput ↓ 49%2h PERFORMANCEGLM-5.2 via OpenRouter: TTFT ↑ 121%2h PERFORMANCEMiniMax M3 via OpenRouter: reliability recovered 96.8% → 100.0%2h PERFORMANCEDeepSeek V4 Pro via Cortecs: throughput ↑ 27%2h PERFORMANCEGPT-OSS 120B via OpenRouter: throughput ↓ 22%2h PERFORMANCEGLM-4.7 via OpenRouter: TTFT ↓ 78%2h PERFORMANCEGPT-OSS 20B → Groq: TTFT ↑ 24%2h PERFORMANCEKimi K3 → Together AI: TTFT ↓ 34%2h PERFORMANCEDeepSeek V4 Flash via OpenRouter: throughput ↓ 17%2h PERFORMANCELlama 3.3 70B via Cortecs: TTFT ↓ 51%2h PERFORMANCEQwen3 235B via OpenRouter: throughput ↑ 37%2h PERFORMANCEQwen3 235B via OpenRouter: TTFT ↓ 29%2h PERFORMANCEDeepSeek V4 Flash → Together AI: throughput ↑ 19%2h PERFORMANCEGLM-5.3 via OpenRouter: throughput ↓ 17%2h PERFORMANCEGPT-OSS 20B via Cortecs: throughput ↑ 28%2h PERFORMANCELlama 3.3 70B → Together AI: TTFT ↓ 33%2h PRICEtencent/hy4-preview output_per_mtok increased from 2.25 to 2.502h

GPT-OSS 120B

11 providers · 117B MoE · 128k context · 2 regions · all models →

GPT-OSS 120B across 11 providers

Same model. Different infrastructure.

TTFT spread2.6×Throughput spread25.7×Output price4.4×Tool reliability0.0 pts

TTFT179 ms474 msbest: Scaleway

Throughput68 tok/s1749 tok/sbest: Cerebras

Output price$0.17/M$0.75/Mbest: OpenRouter

Reliability93.5%100.0%best: Scaleway

Tool success100.0%100.0%best: Scaleway

GPT-OSS 120B — tool-calling benchmark

6 providers · 100.0% → 100.0% · 0.0 pt spread

#	Provider	Tools	Δ best	Rel.	JSON	TTFT	P95	Tok/s	$ Out
01	Cerebras	100.0%	—	100.0%	100.0%	184 ms	323 ms	1749	$0.75
02	Groq	100.0%	−0.0	100.0%	100.0%	265 ms	400 ms	486	$0.60
03	OpenRouter	100.0%	−0.0	99.1%	100.0%	474 ms	1413 ms	68	$0.17
04	Cortecs	100.0%	−0.0	100.0%	100.0%	237 ms	521 ms	351	$0.46
05	Together	100.0%	−0.0	93.5%	85.7%	182 ms	467 ms	138	$0.60
06	Scaleway	100.0%	−0.0	100.0%	100.0%	179 ms	2214 ms	96	$0.69

GPT-OSS 120B inference benchmark: fastest & cheapest providers · InferenceBench