Inference · every season

Every model, one table.

Every model this network has run in 14 weeks of seasons: how fast it ran and on what machine, what a million tokens costs here next to what OpenRouter charges, what it proved it can do, and who its licence lets you serve.

96models measured here
82that no hosted API sells
137sold by OpenRouter
0cheaper here than OpenRouter
what a million tokens costs here

Every machine is metered at one posted price list: $0.036 a CPU-hour, $0.003 a GB of memory an hour, $0.36 a GPU-hour and $0.1 a kWh. A model's price here is that list applied to the machine its signed result names (its CPUs, memory and card) for the time it takes to write a million tokens at its signed speed.

why the API is usually cheaper

A hosted API runs one model for thousands of people on data-centre GPUs. One machine running one model for one person can't match that price. Running here pays off when the work must stay on your own machines, or when no API sells the model at all.

who you may serve

Read from each model's licence. Apache, MIT, Llama and Gemma terms let you serve anyone. Custom or community licences (Qwen's, MiniMax's): your own machines and your company's only. Non-commercial ones: your own machines, unpaid work.

91 modelsclick a column to sort by it · click a model for every machine it ran on
parameterstokens/secprovedper M tokensper M tokensVerdict
Google
Text & chat · by Google18 builds
5.1B2B active3.8 GB173NVIDIANVIDIA 12 GB128k10 + 20 lang$1.39meterednot soldAnyoneonly on this network
Google
Text & chat · by Google8 builds
8B4B active4.6 GB83NVIDIANVIDIA 15 GB128k10 + 14 lang$1.73meterednot soldAnyoneonly on this network
Google
Text & chat · by Google5 builds
12B6.6 GB24NVIDIANVIDIA 15 GB256k10 + 14 lang$5.79meterednot soldAnyoneonly on this network
Google
Text & chat · by Google3 builds
26B4B active15.8 GB18NVIDIANVIDIA 15 GB128k10 + 20 lang$8.79metered$0.3015 providers · free tierAnyoneAPI 29× cheaper
Google
Text & chat · by Google2 builds
31B—2.3NVIDIANVIDIA 15 GB128k10 + 20 lang$57metered$0.4016 providers · free tierAnyoneAPI 144× cheaper
Qwen
Text & chat · by Qwen8 builds
4B2.3 GB90NVIDIANVIDIA 12 GB64k8 + 5 lang$2.21meterednot soldAnyoneonly on this network
Qwen
Text & chat · by Qwen3 builds
600M0.6 GB45machine not reported4k8 + 1 lang—not soldAnyoneonly on this network
Text & chat · by yuxinlu14 builds
12B—26NVIDIANVIDIA 15 GB128k8 + 13 lang$5.21meterednot soldNot checkedonly on this network
Qwen
Text & chat · by Qwen5 builds
4B7.5 GB12NVIDIANVIDIA 15 GB128k8 + 9 lang$12meterednot soldAnyoneonly on this network
Qwen
Text & chat · by Qwen3 builds
14B8.4 GBran, not timed64k8 + 7 lang—$0.242 providersAnyonemeasured
openai
Text & chat · by openai3 builds
20B10.8 GB4.44 CPUs, no GPU·KaggleKaggle4k7 + 9 lang$8.95metered$0.0911 providersAnyoneAPI 99× cheaper
moonshotai
Text & chat · by moonshotai1 build
48B3B active28 GB2.94 CPUs, no GPU4k7 + 9 lang$18meterednot soldAnyoneonly on this network
zai-org
Text & chat · by zai-org3 builds · 1 variant
≈28B15.7 GB2.54 CPUs, no GPU4k7 + 8 lang$18metered$0.402 providersAnyoneAPI 45× cheaper
CohereLabs
Text & chat · by CohereLabs4 builds
≈31B17.2 GB2.24 CPUs, no GPU4k7 + 9 lang$20meterednot soldAnyoneonly on this network
Qwen
Text & chat · by Qwen1 build
1.7B—ran, not timed64k7 + 1 lang—not soldAnyoneonly on this network
Qwen
Text & chat · by Qwen1 build
8B—ran, not timed64k7 + 4 lang—not soldAnyoneonly on this network
meta-llama
Text & chat · by meta-llama16 builds · 2 variants
3.2B1.9 GB225NVIDIANVIDIA 12 GB128k6 + 8 lang$1.14metered$0.332 providersAnyoneAPI 3.5× cheaper
Qwen
Text & chat · by Qwen3 builds
800M0.5 GB164 CPUs, no GPU4k6 + 1 lang$1.91meterednot soldAnyoneonly on this network
LiquidAI
Text & chat · by LiquidAI1 build
1.2B0.7 GB164 CPUs, no GPU4k6 + 1 lang$1.95meterednot soldYou & your companyonly on this network
Qwen
Text & chat · by Qwen2 builds · 1 variant
8B4.7 GBran, not timed128k6 + 7 lang—$0.751 providerAnyonemeasured
microsoft
Text & chat · by microsoft10 builds
≈4.1B2.3 GB189NVIDIANVIDIA 12 GB4k5 + 3 lang$1.36meterednot soldAnyoneonly on this network
meta-llama
Text & chat · by meta-llama11 builds
1.2B0.8 GB20AppleApple silicon32k5 + 8 lang$2.66metered$0.201 providerAnyoneAPI 13× cheaper
tencent
Text & chat · by tencent2 builds · 1 variant
30B3B active17 GB3.64 CPUs, no GPU4k5 + 10 lang$12metered$0.291 providerAnyoneAPI 41× cheaper
redhatai
Text & chat · by redhatai1 build
——ran, not timed32k5 + 6 lang—not soldNot checkedonly on this network
Qwen
Text & chat · by Qwen3 builds
500M0.5 GB239Linux · NVIDIA · 32 GB32k4 + 5 lang$0.13meterednot soldAnyoneonly on this network
microsoft
Text & chat · by microsoft20 builds
≈4.1B2.3 GB154NVIDIANVIDIA 12 GB128k4 + 5 lang$1.66meterednot soldAnyoneonly on this network
google
Text & chat · by google13 builds · 2 variants
270M0.2 GB404 CPUs, no GPU—4 + 3 lang$0.76meterednot soldAnyoneonly on this network
Qwen
Text & chat · by Qwen2 builds
30B3B active16 GB27NVIDIANVIDIA 15 GB128k4 + 9 lang$5.05metered$0.282 providersAnyoneAPI 18× cheaper
sarvamai
Text & chat · by sarvamai1 build
24B13.3 GB16NVIDIANVIDIA 15 GB8k4 + 9 lang$9.33meterednot soldAnyoneonly on this network
krutrim-ai-labs
Text & chat · by krutrim-ai-labs1 build
12B7 GB14NVIDIANVIDIA 15 GB64k4 + 9 lang$10meterednot soldYou & your companyonly on this network
ibm-granite
Text & chat · by ibm-granite6 builds
1B1 GB114 CPUs, no GPU4k4 + 6 lang$2.85meterednot soldAnyoneonly on this network
microsoft
Text & chat · by microsoft3 builds
≈15B8.4 GB3.3AppleApple silicon32k4 + 9 lang$60meterednot soldAnyoneonly on this network
google
Text & chat · by google7 builds
270M0.2 GB374 CPUs, no GPU32k3 + 2 lang$0.82meterednot soldAnyoneonly on this network
google
Text & chat · by google6 builds
1B0.8 GB134 CPUs, no GPU8k3 + 8 lang$2.45meterednot soldAnyoneonly on this network
sarvamai
Text & chat · by sarvamai2 builds
2B1.4 GB111NVIDIANVIDIA 15 GB8k2 + 8 lang$1.22meterednot soldNot checkedonly on this network
mradermacher
Text & chat · by mradermacher7 builds
270M—394 CPUs, no GPU—2 + 3 lang$0.79meterednot soldNot checkedonly on this network
ai4bharat
Text & chat · by ai4bharat1 build
7B3.9 GB32NVIDIANVIDIA 15 GB—1 + 2 lang$4.36meterednot soldYou & your companyonly on this network
google
Embeddings (search) · by google13 builds
300M0.2 GBran, not timed—1—not soldAnyoneonly on this network
SWivid
Text to speech · by SWivid1 build
150M0.6 GBran, not timed—1—not soldOwn machines onlyonly on this network
Text to speech · by hexgrad1 build
—0.3 GBran, not timed—1 + 1 lang—not soldNot checkedonly on this network
facebook
Text to speech · by facebook1 build
40M0.1 GBran, not timed—1—not soldOwn machines onlyonly on this network
facebook
Text to speech · by facebook1 build
40M0.1 GBran, not timed—1—not soldOwn machines onlyonly on this network
facebook
Text to speech · by facebook1 build
40M0.1 GBran, not timed—1—not soldOwn machines onlyonly on this network
facebook
Text to speech · by facebook1 build
40M0.1 GBran, not timed—1—not soldOwn machines onlyonly on this network
facebook
Text to speech · by facebook1 build
40M0.1 GBran, not timed—1—not soldOwn machines onlyonly on this network
facebook
Text to speech · by facebook1 build
40M0.1 GBran, not timed—1—not soldOwn machines onlyonly on this network
facebook
Text to speech · by facebook1 build
40M0.1 GBran, not timed—1—not soldOwn machines onlyonly on this network
facebook
Text to speech · by facebook1 build
40M0.1 GBran, not timed—1—not soldOwn machines onlyonly on this network
facebook
Text to speech · by facebook1 build
40M0.1 GBran, not timed—1—not soldOwn machines onlyonly on this network
facebook
Text to speech · by facebook1 build
40M0.1 GBran, not timed—1—not soldOwn machines onlyonly on this network
facebook
Text to speech · by facebook1 build
40M0.1 GBran, not timed—1—not soldOwn machines onlyonly on this network
facebook
Text to speech · by facebook1 build
40M0.1 GBran, not timed—1—not soldOwn machines onlyonly on this network
Speech to text · by UsefulSensors1 build
—0.2 GBran, not timed—1 + 1 lang—not soldAnyoneonly on this network
Speech to text · by UsefulSensors1 build
—0.1 GBran, not timed—1 + 1 lang—not soldAnyoneonly on this network
facebook
Music & sound · by facebook1 build
1.5B11.1 GBran, not timed—1—not soldOwn machines onlyonly on this network
facebook
Music & sound · by facebook1 build
300M2.2 GBran, not timed—1—not soldOwn machines onlyonly on this network
Other · by rhasspy1 build
——ran, not timed—1 + 1 lang—not soldNot checkedonly on this network
facebook
Vision (counting, locating) · by facebook1 build
90M0.7 GBran, not timed—1—not soldAnyoneonly on this network
Efficient-Large-Model
Image generation · by Efficient-Large-Model1 build
1.6B3 GBran, not timed—1—not soldAnyoneonly on this network
stabilityai
Image generation · by stabilityai1 build
900M2.4 GBran, not timed—1—not soldNot checkedonly on this network
speechbrain
Speaker recognition · by speechbrain1 build
20M0.1 GBran, not timed—1—not soldAnyoneonly on this network
Wan-AI
Video generation · by Wan-AI1 build
1.3B26.9 GBran, not timed—1—not soldAnyoneonly on this network
segmind
Image generation · by segmind1 build
320M1 GBran, not timed—1—not soldAnyoneonly on this network
openai
Speech to text · by openai1 build
810M1.5 GBran, not timed—1 + 1 lang—not soldAnyoneonly on this network
openai
Speech to text · by openai1 build
240M0.9 GBran, not timed—1 + 6 lang—not soldAnyoneonly on this network
vasista22
Speech to text · by vasista221 build
240M0.9 GBran, not timed—1 + 1 lang—not soldAnyoneonly on this network
vasista22
Speech to text · by vasista221 build
240M0.9 GBran, not timed—1 + 1 lang—not soldAnyoneonly on this network
vasista22
Speech to text · by vasista221 build
240M0.9 GBran, not timed—1 + 1 lang—not soldAnyoneonly on this network
tencent
Text & chat · by tencent1 build
1.8B1 GB132NVIDIA2× Tesla T4 15 GB·KaggleKaggle——$1.03meterednot soldNot checkedonly on this network
sarvamai
Text & chat · by sarvamai1 build
7B3.9 GB30NVIDIANVIDIA 15 GB——$4.62meterednot soldAnyoneonly on this network
meta-models
Text & chat · by meta-models5 builds · 1 variant
30B15.6 GB1.3machine not reported———$1.202 providersAnyonemeasured
Qwen
Text & chat · by Qwen3 builds
27B15 GB0.8machine not reported———$3.207 providersAnyonemeasured
THUDM
Image generation · by THUDM1 build
5B20 GBran, not timed———not soldYou & your companyonly on this network
THUDM
Image generation · by THUDM2 builds
2B12.8 GBran, not timed———not soldAnyoneonly on this network
depth-anything
Vision (counting, locating) · by depth-anything1 build
100M0.4 GBran, not timed———not soldOwn machines onlyonly on this network
depth-anything
Vision (counting, locating) · by depth-anything1 build
30M0.1 GBran, not timed———not soldAnyoneonly on this network
tencent
Image generation · by tencent1 build
13B39 GBran, not timed———not soldNot checkedonly on this network
Lightricks
Image generation · by Lightricks1 build
2B23.6 GBran, not timed———not soldYou & your companyonly on this network
lightricks
Other · by lightricks1 build
13B—ran, not timed———not soldNot checkedonly on this network
google
Translation · by google1 build
3B11 GBran, not timed—0 + 10 lang—not soldAnyoneonly on this network
facebook
Translation · by facebook1 build
1.3B5.1 GBran, not timed—0 + 12 lang—not soldOwn machines onlyonly on this network
facebook
Translation · by facebook1 build
600M2.3 GBran, not timed—0 + 12 lang—not soldOwn machines onlyonly on this network
Other · by prunaai1 build
——ran, not timed———not soldNot checkedonly on this network
Other · by harahan1 build
2B—ran, not timed———not soldNot checkedonly on this network
Other · by lightx2v1 build
——ran, not timed———not soldNot checkedonly on this network
sarvamai
Translation · by sarvamai1 build
4B8 GBran, not timed—0 + 12 lang—$0.102 providersAnyonemeasured
Other · by languaman1 build
——ran, not timed———not soldNot checkedonly on this network
Other · by aemotionstudio1 build
——ran, not timed———not soldNot checkedonly on this network
optimum-intel-internal-testing
Other · by optimum-intel-internal-testing1 build
——ran, not timed———not soldNot checkedonly on this network
Other · by alibaba-pai1 build
1.3B—ran, not timed———not soldNot checkedonly on this network
wan-ai
Other · by wan-ai1 build
1.3B—ran, not timed———not soldNot checkedonly on this network

Updated Wed, 28 Oct 2026 06:31:54 GMT. The data: table.json (models, and every build behind them), market.json (OpenRouter prices and licences, with when they were read). Back to the seasons →