model prices compared · prices as of · 97 models both sell
DeepInfra is cheaper on 26 of the 97 models both sell, OpenRouter on 24, and 47 cost the same at both. DeepInfra is a host and OpenRouter a router, so the two catalogs overlap without matching. The widest gap is Llama 3.2 3B Instruct, $0.08 per million tokens at DeepInfra against $0.48.
DeepInfra is cheaper on 26 of the 97 models both sell, and the cheapest thing either one offers is Llama 3.2 3B Instruct at $0.08 per million tokens, read 16 Sept 2026.
The reference router: charges each host's own price and routes open-weight models to the cheapest healthy host, so its price for an open model is often below the maker's.
Most widely sold first. Each price is the cheapest that provider charges for the model under the standard blend, three parts input to one part output per million tokens. A free tier is a limit rather than a price, and a price below the maker's own with nothing published to account for it is left out, so neither can be marked as the cheaper of the two.
ModelDeepInfraOpenRouter
DeepSeek V4 Flash 0731DeepSeek · 1M context$0.45$0.09 / $0.18 in / out$0.22$0.04 / $0.10 in / out
GLM-5.2Z.ai · 1M context$4.65$0.75 / $2.40 in / out$8.60$1.40 / $4.40 in / out
DeepSeek V4 ProDeepSeek · 1M context$6.50$1.30 / $2.60 in / out$5.90$0.98 / $2.95 in / out
Kimi K2.6Moonshot AI · 262k context$5.75$0.75 / $3.50 in / out$6.85$0.95 / $4 in / out
Kimi K3Moonshot AI · 1M context$22.80$2.85 / $14.25 in / out$21.23$2.65 / $13.28 in / out
GLM-5.3-FlashZ.ai · 1M context$0.95$0.15 / $0.50 in / out$0.57$0.09 / $0.30 in / out
GLM-5.3Z.ai · 1M context$7.60$1.20 / $4 in / out$8.60$1.40 / $4.40 in / out
Kimi K2.7 CodeMoonshot AI · 262k context$5.44$0.68 / $3.40 in / out$5.33$0.71 / $3.21 in / out
gpt-oss-120bOpenAI · 131k context$0.28$0.04 / $0.17 in / out$0.28$0.04 / $0.17 in / out
GLM-5.1Z.ai · 200k context$6.65$1.05 / $3.50 in / out$5.93$0.97 / $3.04 in / out
MiniMax-M3MiniMax · 1M context$1.94$0.28 / $1.10 in / out$2.10$0.30 / $1.20 in / out
DeepSeek V3.2DeepSeek · 164k context$1.16$0.26 / $0.38 in / out$1.21$0.27 / $0.40 in / out
DeepSeek V4.1 FlashDeepSeek · 1M context$1.20$0.20 / $0.60 in / out$1.05$0.15 / $0.60 in / out
Kimi K2.5Moonshot AI · 262k context$3.60$0.45 / $2.25 in / out$3.60$0.45 / $2.25 in / out
Claude Sonnet 4.6Anthropic · 1M context$24$3 / $15 in / out$24$3 / $15 in / out
GLM-5Z.ai · 205k context$3.88$0.60 / $2.08 in / out$3.72$0.60 / $1.92 in / out
Qwen3.8 27BAlibaba Cloud · 1M context$2.33$0.15 / $1.88 in / out$3.19$0.21 / $2.55 in / out
Claude Opus 4.8Anthropic · 1M context$40$5 / $25 in / out$40$5 / $25 in / out
MiniMax-M2.7MiniMax · 205k context$1.75$0.25 / $1 in / out$2.10$0.30 / $1.20 in / out
Claude Haiku 4.5Anthropic · 200k context$8$1 / $5 in / out$8$1 / $5 in / out
Claude Opus 4.7Anthropic · 1M context$40$5 / $25 in / out$40$5 / $25 in / out
MiniMax-M2.5MiniMax · 205k context$1.60$0.15 / $1.15 in / out$1.89$0.27 / $1.08 in / out
Llama-3.3-70B-InstructMeta · 128k context$0.62$0.10 / $0.32 in / out$0.62$0.10 / $0.32 in / out
Qwen3.5 397B-A17BAlibaba Cloud · 262k context$4.35$0.45 / $3 in / out$5.15$0.55 / $3.50 in / out
Claude Fable 5Anthropic · 1M context$80$10 / $50 in / out$80$10 / $50 in / out
gpt-oss-20bOpenAI · 131k context$0.23$0.03 / $0.14 in / out$0.22$0.03 / $0.13 in / out
Claude Sonnet 5Anthropic · 1M context$24$3 / $15 in / out$16$2 / $10 in / out
GLM-4.7Z.ai · 205k context$2.95$0.40 / $1.75 in / out$2.95$0.40 / $1.75 in / out
Claude Opus 5Anthropic · 1M context$40$5 / $25 in / out$40$5 / $25 in / out
On the 97 models both sell, DeepInfra is cheaper on 26, OpenRouter on 24, and 47 cost the same at both. The widest gap is Llama 3.2 3B Instruct: $0.08 per million tokens against $0.48. Prices are blended three parts input to one part output and read on 16 Sept 2026.
Which models do DeepInfra and OpenRouter both sell?
97 of the models we track, led by DeepSeek V4 Flash 0731, GLM-5.2, DeepSeek V4 Pro. DeepInfra sells 106 in all and OpenRouter 261, so most of each catalog has no counterpart at the other.
Does DeepInfra or OpenRouter train on what you send it?
DeepInfra does not say train on what you send it, and not stated. OpenRouter does not say, and not stated. A provider that may train on prompts is often how a price below the maker's own is paid for.
How current are these prices?
Both catalogs are read every day. DeepInfra was last read 16 Sept 2026 and OpenRouter 16 Sept 2026; the whole index was refreshed 16 Sept 2026. Prices exclude tax, and a price below the maker's own with nothing published to account for it is left out of this table entirely.
DeepInfraall 106 models →
OpenRouterall 261 models →