Inference provider

DeepInfra: AI models and API pricing

Every tracked model DeepInfra serves through OpenRouter, with DeepInfra's own input and output price per 1M tokens, the context window it serves, disclosed quantization and measured endpoint uptime. Rows link to the model page, where every provider's deployment is listed side by side.

Snapshot as of 11 August 2026

DeepInfra serves 51 tracked models through 52 endpoints as of 11 August 2026; the highest-ranked is Kimi K3 (ModelCap Index #4); mean measured uptime is 95.75%.

Models served

On the ModelCap Index

Cheapest output price

$0.03/1M

Language models on DeepInfra (51)

Ranked models first, in ModelCap Index order. Prices are the lowest DeepInfra lists for the model; the endpoint count shows how many deployments carry it.

Language models served by DeepInfra, with price, context, quantization and uptime
#	Model	ModelCap Index	Input $/1M	Output $/1M	Context	Quantization	Uptime (30 min)	Endpoints
1	Kimi K3Moonshot AI	#478.4	$2.85	$14.25	1M	bf16	65.77%	1
2	GLM 5.2Z.ai	#1370.5	$0.75	$2.40	1M	fp4	88.70%	1Open weights
3	Hy3Tencent	#1864.9	$0.14	$0.58	262K	fp8	100.0%	1Open weights
4	Gemma 4 31BGoogle	#2062.8	$0.09	$0.34	262K	fp4, fp8	99.79%	2Open weights
5	Qwen3.5 397B A17BQwen	#2261.6	$0.45	$3.00	262K	fp8	99.76%	1Open weights
6	MiniMax M3MiniMax	#2361.3	$0.28	$1.10	524K	fp8	50.57%	1
7	InklingThinking Machines	#2460.7	$0.95	$4.05	524K	fp8	100.0%	1Open weights
8	Gemma 4 26B A4BGoogle	#2559.2	$0.07	$0.34	262K	fp8	99.76%	1Open weights
9	Qwen3 235B A22B Instruct 2507Qwen	#3153.6	$0.09	$0.55	262K	fp8	99.67%	1Open weights
10	Qwen3.6 27BQwen	#3251.9	$0.32	$3.20	262K	fp8	88.24%	1Open weights
11	Qwen3.5-122B-A10BQwen	#3351.8	$0.29	$2.40	262K	fp4	90.55%	1Open weights
12	Qwen3 VL 235B A22B InstructQwen	#3451.5	$0.20	$0.88	262K	fp8	—	1Open weights
13	Qwen3 Next 80B A3B InstructQwen	#4046.6	$0.09	$1.10	262K	fp8	100.0%	1Open weights
14	Qwen3.6 35B A3BQwen	#4146.3	$0.10	$0.95	262K	fp8	100.0%	1Open weights
15	Qwen3 235B A22B Thinking 2507Qwen	#4345.3	$0.23	$2.30	262K	fp8	—	1Open weights
16	Step 3.7 FlashStepFun	#4445.3	$0.20	$1.15	262K	—	100.0%	1Open weights
17	Qwen3 Coder 480B A35BQwen	#4643.7	$0.30	$1.00	262K	fp4	70.31%	1Open weights
18	GLM 4.7 FlashZ.ai	#6236.2	$0.06	$0.40	203K	bf16	99.56%	1Open weights
19	Gemma 3 27BGoogle	#6533.9	$0.08	$0.16	131K	fp8	99.70%	1
20	Ling-3.0-flashInclusionAI	#6633.4	$0.06	$0.18	131K	bf16	97.84%	1Open weights
21	Nemotron 3 SuperNVIDIA	#6733.0	$0.085	$0.40	262K	bf16	89.52%	1
22	Mistral Small 3.2 24BMistral AI	#6832.3	$0.075	$0.20	128K	fp8	99.90%	1Open weights
23	Qwen3 32BQwen	#7229.2	$0.08	$0.28	41K	fp8	99.94%	1Open weights
24	Qwen3 14BQwen	#8725.7	$0.12	$0.24	41K	fp8	100.0%	1Open weights
25	Gemma 3 12BGoogle	#9325.5	$0.05	$0.15	131K	bf16	100.0%	1
26	Qwen3 30B A3BQwen	#18024.3	$0.12	$0.50	41K	fp8	100.0%	1Open weights
27	Llama 4 MaverickMeta	#18123.5	$0.20	$0.80	1M	fp8	99.94%	1
28	Llama 4 ScoutMeta	#18821.7	$0.10	$0.30	328K	fp8	99.82%	1
29	Kimi K2.7 CodeMoonshot AI	#18921.5	$0.68	$3.40	262K	fp4	100.0%	1
30	Inkling SmallThinking Machines	#19220.4	$0.45	$1.20	524K	fp8	90.05%	1Open weights
31	Llama 3.3 70B InstructMeta	#19320.3	$0.10	$0.32	131K	fp8	97.47%	1
32	Nemotron 3 Nano 30B A3BNVIDIA	#19519.9	$0.05	$0.20	262K	fp4	90.43%	1
33	Qwen2.5 72B InstructQwen	#19816.9	$0.36	$0.40	33K	fp8	99.89%	1
34	Gemma 3 4BGoogle	#20014.4	$0.05	$0.10	131K	bf16	100.0%	1
35	Mistral NemoMistral AI	#20311.9	$0.019	$0.03	131K	fp8	99.34%	1Open weights
36	Phi 4Microsoft	#2089.2	$0.07	$0.14	16K	bf16	100.0%	1Open weights
37	Qwen3 VL 30B A3B InstructQwen	#2184.1	$0.15	$0.60	262K	fp8	99.70%	1Open weights
38	Llama 3.1 8B InstructMeta	#2193.7	$0.02	$0.04	131K	fp8	98.51%	1
39	Nemotron 3 UltraNVIDIA	#2202.7	$0.50	$2.20	262K	fp4	—	1
40	Llama Guard 4 12BMeta	#2290.0	$0.18	$0.18	164K	bf16	99.84%	1
41	Qwen3.5-9BQwen	—	$0.10	$0.15	262K	bf16	99.52%	1Open weights
42	GLM 4.6Z.ai	—	$0.50	$2.00	203K	fp4	99.49%	1Open weights
43	GLM 4.7Z.ai	—	$0.40	$1.75	203K	fp4	100.0%	1Open weights
44	GLM 5Z.ai	—	$0.60	$2.08	203K	fp4	95.93%	1Open weights
45	GLM 5.1Z.ai	—	$1.05	$3.50	203K	fp4	100.0%	1Open weights
46	Kimi K2.5Moonshot AI	—	$0.45	$2.25	262K	fp4	99.94%	1
47	Kimi K2.6Moonshot AI	—	$0.75	$3.50	262K	fp4	88.33%	1
48	Llama 3.1 70B InstructMeta	—	$0.40	$0.40	131K	fp8	99.83%	1
49	Mistral Small 3Mistral AI	—	$0.05	$0.08	33K	fp8	100.0%	1Open weights
50	Qwen3.5-27BQwen	—	$0.26	$2.60	262K	fp8	98.75%	1Open weights
51	Qwen3.5-35B-A3BQwen	—	$0.14	$1.00	262K	fp8	—	1Open weights

DeepInfra on ModelCap: common questions

How many AI models does DeepInfra serve?

51 tracked models as of 11 August 2026: 51 language models (40 with a public ModelCap Index position), across 52 live endpoints observed through OpenRouter.

What is the best model available on DeepInfra?

Kimi K3 from Moonshot AI is the highest-ranked language model DeepInfra serves, at ModelCap Index position #4 and $14.25 per 1M output tokens on DeepInfra. The table lists every model in Index order.

What is the cheapest model on DeepInfra?

Mistral Nemo at $0.03 per 1M output tokens ($0.019 per 1M input), the lowest positive language-token price DeepInfra lists. Free-tier endpoints, where present, show as $0 and are excluded from that comparison.

Which DeepInfra model has the largest context window?

Kimi K3, served with a 1M-token context window on DeepInfra. Providers sometimes serve less than the model's published maximum, so the figure here is the provider's own.

How reliable is DeepInfra?

Mean measured uptime across DeepInfra endpoints is 95.75% over the last 30 minutes, as reported by OpenRouter at the time of the snapshot. Per-model uptime is in the table; it is a live operational signal, not a service-level guarantee.

Where do the DeepInfra prices come from?

From the OpenRouter catalogue entry for each DeepInfra endpoint, in USD per 1M tokens, re-observed on every ModelCap refresh. Where DeepInfra exposes several deployments of one model at different prices, the table shows the lowest and notes the count; the model page lists each deployment.

How current is this page?

It re-renders within a minute of every dataset refresh; this snapshot was published 11 August 2026. Offers retained from a failed refresh are marked stale in the data and are not counted here.