Same model, different price
The model you choose is not the price you pay. 112 models in this index run on two or more hosts, 2 to 32 of them each, and those hosts do not agree: the median gap between the cheapest and the priciest is 62%. Same weights, a different bill.
deepseek-v3.2 is served by 14 providers. Cheapest: $0.518/1M blended at GMICloud. Priciest: $7.50 at SambaNova, 14.5×. Routing to the priciest host instead of the cheapest costs $4.89 more per 1,000 chatbot requests.
The widest gaps, drawn
One row per model, from its cheapest host (hollow) to its priciest (solid). Models on three or more hosts, widest first; the segment's length is the decision you make without being asked.
Cheapest → priciest host
blended USD per 1M tokens (1,000 in / 500 out weighting), log scale
◦ a promotional rate, not the list price · The 20 widest of 83 models on three or more hosts.
Every model served by more than one host
Widest gap first. Waste = the extra cost per 1,000 chatbot requests (1,000 in / 500 out tokens) if you take the priciest host over the cheapest.
1 of these 112 gaps rest on a price our sources disagree about, marked †. We publish the gap anyway, since hiding the row would hide the disagreement too, but a dagger means that end is soft and the percentage could move a lot once the sources are reconciled. How the flag is assigned.
25 of these 112 gaps have an end that is a promotional rate, marked promo. We rank on what you pay today, but a promotion ends at the list price, and a gap that rests on one narrows when it does. The list price and the discount are on the endpoint's own page.
| deepseek-v3.2 | $0.518 GMICloud promo | 14.5× |
| glm-5.3-flash | $0.189 StreamLake promo | 13.2× |
| mimo-v2.6-pro | $1.17 DeepInfra promo | 11.2× |
| deepseek-v4-flash | $0.21 Relace | 8.4× |
| deepseek-v4-flash-0423 | $0.21 StreamLake promo | 8.4× |
| qwen3-coder-30b-a3b | $0.34 Novita AI | 7.9× |
| mistral-nemo | $0.038 Io Net promo | 7.8× |
| llama-3.1-8b | $0.06 DeepInfra | 7.3× |
| gpt-oss-120b | $0.18 Venice | 6.1× |
| llama-3.3-70b | $0.42 DeepInfra | 6.1× |
| mimo-v2.5 | $0.399 DeepInfra promo | 6.0× |
| nemotron-3-super-120b-a12b | $0.485 DeepInfra | 5.6× |
| qwen3-14b | $0.32 NextBit | 5.5× |
| gemma-4-31b | $0.38 Reka | 4.6× |
| qwen3-coder | $1.30 DeepInfra | 4.5× |
| deepseek-v4.1-flash-1017 | $0.392 DeepInfra promo | 4.2× |
| mimo-v2.5-pro | $0.913 GMICloud promo | 4.2× |
| ling-3.1-flash | $0.3 DeepInfra promo | 4.0× |
| qwen3.8-27b | $1.29 Chutes | 3.7× |
| deepseek-v4.1-flash | $0.468 Io Net promo | 3.5× |
| glm-5.2 | $2.36 DeepInfra promo | 3.5× |
| gpt-oss-20b | $0.12 AkashML | 3.1× |
| nemotron-3.5-super-vl | $0.5 Crusoe | 3.0× |
| glm-5.3 | $2.36 Inceptron | +195% |
| gemma-4-26b-a4b | $0.26 Reka | +188% |
| minimax-m2.7 | $1.05 GMICloud promo | +186% |
| qwen3-30b-a3b | $0.241 StreamLake promo | +170% |
| deepseek-v4 | $2.50 Relace | +164% |
| minimax-m3 | $1.19 CoreWeave | +152% |
| qwen3-235b-a22b | $0.438 GMICloud promo | +151% |
| deepseek-v4-pro-0423 | $2.40 Relace | +139% |
| llama-4-scout | $0.4 DeepInfra | +137% |
| kimi-k2.6 | $2.40 Chutes | +137% |
| gemma-3-27b | $0.24 DeepInfra | +121% |
| qwen3.6-35b-a3b | $1.00 AkashML | +104% |
| deepseek-v4-flash-vision-exp | $0.862 DeepInfra promo | +104% |
| deepseek-v3.2-exp | $0.336 AtlasCloud | +102% |
| qwen3-vl-30b-a3b-thinking | $1.29 SiliconFlow | +102% |
| llama-3.1-70b | $0.8 DeepInfra | +100% |
| qwen3.7-max | $5.00 Novita AI | +100% |
| qwen3.5-35b-a3b | $1.14 DeepInfra | +97% |
| qwen3-32b | $0.36 DeepInfra | +97% |
| qwen3-coder-next | $0.92 Parasail | +96% |
| qwen3-vl-235b-a22b | $1.08 DeepInfra | +95% |
| qwen3-next-80b-a3b | $0.878 Alibaba † | +88% |
| qwen3.6-27b | $2.25 Chutes | +87% |
| mimo-v2.6-flash | $0.28 Relace | +87% |
| deepseek-chat-v3.1 | $1.20 DeepInfra | +83% |
| gemma-3-12b | $0.15 Novita AI | +83% |
| kimi-k3 | $11.35 Chutes | +82% |
| deepseek-chat-v3 | $1.14 DeepInfra | +75% |
| qwen3-vl-8b | $0.572 Alibaba | +75% |
| qwen3-vl-30b-a3b | $0.75 DeepInfra | +72% |
| qwen3.5-9b | $0.25 SiliconFlow | +68% |
| glm-5 | $2.52 GMICloud promo | +67% |
| gemma-3-4b | $0.09 Parasail | +67% |
| qwen3-235b-a22b-thinking | $2.53 Alibaba | +56% |
| qwen3.5-122b-a10b | $2.34 SiliconFlow | +54% |
| qwen3.5-397b-a17b | $3.45 DeepInfra | +52% |
| llama-4-maverick | $1.00 DeepInfra | +50% |
| nemotron-3.5-lightning | $0.181 Io Net promo | +49% |
| qwen3-max | $7.20 Alibaba | +47% |
| glm-5.1 | $4.00 StreamLake promo | +45% |
| mistral-small-3.2-24b | $0.275 DeepInfra | +45% |
| kimi-k2.5 | $2.70 DeepInfra | +43% |
| nemotron-3-ultra-550b-a55b | $2.70 DeepInfra | +39% |
| kimi-k2.7-code | $3.71 StreamLake promo | +33% |
| glm-4.5-air | $0.98 Novita AI | +33% |
| glm-4.7 | $2.15 DeepInfra | +30% |
| glm-4.6 | $2.18 Venice | +28% |
| qwen3.5-27b | $2.25 SiliconFlow | +27% |
| mistral-small-4 | $0.75 Mistral AI | +25% |
| minimax-m3.1-flash | $0.6 InferenceNet | +25% |
| granite-4.2-8b | $0.25 CoreWeave | +24% |
| minimax-m2.5 | $1.22 Venice | +23% |
| muse-glimmer-30b | $1.50 DeepInfra | +23% |
| ling-3.0-flash-vl | $0.24 DeepInfra | +23% |
| deepseek-r1 | $2.65 DeepInfra | +21% |
| nemotron-3-nano-30b-a3b | $0.25 Crusoe | +20% |
| minimax-m2 | $1.27 Minimax promo | +18% |
| qwen3-vl-235b-a22b-thinking | $4.40 Alibaba | +12% |
| grok-4.6 | $8.00 xAI | +10% |
| grok-4.5 | $8.00 xAI | +10% |
| gpt-5.6-sol | $24.00 OpenAI | +10% |
| gpt-5.6-terra | $14.00 OpenAI | +10% |
| gpt-5.6-luna | $1.40 OpenAI | +10% |
| hy3 | $0.66 DeepInfra | +9% |
| deepseek-v3.1-terminus | $1.27 SiliconFlow | +8% |
| deepseek-chat | $1.21 DeepInfra | +6% |
| minimax-m1 | $2.60 Minimax | +6% |
| deepseek-v3 | $1.21 DeepInfra | +6% |
| qwen-2.5-72b | $0.76 DeepInfra | +3% |
| inkling-small | $1.65 DeepInfra | +3% |
| glm-4.7-flash | $0.46 Venice | +2% |
| llama-3.2-3b | $0.38 Parasail | +2% |
| inkling | $5.00 DeepInfra | +1% |
| claude-opus-5 | $30.00 Anthropic | +0% |
| claude-sonnet-5 | $12.00 Anthropic | +0% |
| claude-fable-5 | $60.00 Anthropic | +0% |
| claude-fable-5-1 | $60.00 Anthropic | +0% |
| qwen3.8-2.4t-a95b | $8.00 Novita AI | +0% |
| minimax-m2.1 | $1.50 Novita AI | +0% |
| glm-4.6v | $1.20 Novita AI | +0% |
| kimi-k2-thinking | $3.10 Google | +0% |
| qwen3-next-80b-a3b-thinking | $1.35 Google | +0% |
| glm-4.5v | $2.40 Novita AI | +0% |
| qwen3.8-flash | $0.62 Alibaba | +0% |
| ling-3.0-flash | $0.24 Novita AI | +0% |
| qwen3.8-max | $8.00 Novita AI | +0% |
| glm-5v-turbo | $5.20 Z.AI | +0% |
| step-5-preview | $3.70 StepFun | +0% |
| muse-spark-open | $5.50 Together AI | +0% |
Not the same product
A cheaper host can serve a quantized build, cap context, throttle harder or sit far from your users. None of that is in a price list. Speed and uptime
Prices
Each provider's own published rates, cross-checked against public price indexes; a disagreement is flagged, never smoothed. Methodology
Start, not end
Take the cheapest host as the start of a decision. The value ranking · The full catalogue
FAQ
Does the same model cost the same on every provider?
No, and the gap is not small. Across 112 models hosted by two or more providers, the median gap between the cheapest and priciest host is 62%, reaching 14.5× at the extreme. The weights are the same artifact; the bill is not.
How much can picking the wrong provider cost?
For deepseek-v3.2, served by 14 providers, the priciest host is 14.5× the cheapest — about $4.89 extra per 1,000 chatbot requests (1,000 in / 500 out tokens).
Why do prices differ if the weights are identical?
Hardware, margins, batching, and serving choices — quantization, context limits, rate limits, region. The model is the same artifact, but the service around it is not standardised, which is exactly why the cheapest host is a starting point for a decision rather than the decision itself.
Is the cheapest host always the right pick?
No. A cheaper host can serve a quantized build, cap context below the model's nominal window, throttle harder, or sit in a region far from your users. That is why each model page shows what each host declares it serves (precision, context, features) and its uptime, rather than ranking on price alone.