FastGPU is an MCP server published by co.fastgpu, hosted at fastgpu.co. When last checked on 2026-11-22 it was working: it answers and lists its tools. It declares 2 tools, including list_gpu_prices, match_workload.
Compare live GPU cloud rental prices and match workloads to the cheapest provider.
531 connections through 4 shared values. Tap any node.
Inferred from the host name and headers: what a server runs on, not who runs it.
One entry per GPU model with its cheapest live on-demand cloud rental price in USD per GPU-hour, the provider offering it, and how many providers rent it, from FastGPU's live price feed across marketplaces, neoclouds and hyperscalers (RunPod, Vast.ai, Lambda, AWS and more). Use it to compare GPU rental prices or to find where a named GPU is cheapest to rent. Each price is the provider's own per-GPU rate: cheapest_min_gpu_count says when it is only sold as a multi-GPU instance, and cheapest_billed_separately names what the provider bills on top of it (CPU, memory, storage), null when the rate includes CPU and memory. page_url links to the GPU's live price page with every offer. Rental prices only: it does not rent, reserve or launch GPUs. No key required.
| Parameter | Type | Description |
|---|---|---|
| tier | string | Return only one class of GPU (case-insensitive; combines with vendor). flagship: the current top training and inference accelerators (H100, H200, B200, GB200, MI300X, MI355X). datacenter: earlier or mid-range data-center and workstation cards (A100, L40S, L4, A10, RTX 6000 Ada, V100, MI210). prosume… |
| vendor | string |
Squares are this server's values; dots are other servers sharing them. A shared address or publisher is a lead, not proof of a common operator.
| Return only GPUs made by this vendor (case-insensitive). Combines with tier: both must match. Omit for every vendor. |
{
"name": "list_gpu_prices",
"title": "List cheapest GPU prices",
"description": "One entry per GPU model with its cheapest live on-demand cloud rental price in USD per GPU-hour, the provider offering it, and how many providers rent it, from FastGPU's live price feed across marketplaces, neoclouds and hyperscalers (RunPod, Vast.ai, Lambda, AWS and more). Use it to compare GPU rental prices or to find where a named GPU is cheapest to rent. Each price is the provider's own per-GPU rate: cheapest_min_gpu_count says when it is only sold as a multi-GPU instance, and cheapest_billed_separately names what the provider bills on top of it (CPU, memory, storage), null when the rate includes CPU and memory. page_url links to the GPU's live price page with every offer. Rental prices only: it does not rent, reserve or launch GPUs. No key required.",
"inputSchema": {
"type": "object",
"properties": {
"tier": {
"enum": [
"flagship",
"datacenter",
"prosumer",
"entry"
],
"type": "string",
"description": "Return only one class of GPU (case-insensitive; combines with vendor). flagship: the current top training and inference accelerators (H100, H200, B200, GB200, MI300X, MI355X). datacenter: earlier or mid-range data-center and workstation cards (A100, L40S, L4, A10, RTX 6000 Ada, V100, MI210). prosumer: desktop consumer and workstation cards (RTX 5090, RTX 4090, RTX 3090, RTX A4000). entry: small inference cards (T4, A2, A16, P4). Omit for every tier."
},
"vendor": {
"enum": [
"NVIDIA",
"AMD"
],
"type": "string",
"description": "Return only GPUs made by this vendor (case-insensitive). Combines with tier: both must match. Omit for every vendor."
}
}
},
"outputSchema": {
"type": "object",
"required": [
"count",
"gpus"
],
"properties": {
"gpus": {
"type": "array",
"items": {
"type": "object",
"properties": {
"gpu": {
"type": "string"
},
"url": {
"type": "string",
"description": "Site-relative path of this GPU's live price page, e.g. /gpus/h100-sxm."
},
"arch": {
"type": [
"string",
"null"
]
},
"slug": {
"type": "string"
},
"tier": {
"type": [
"string",
"null"
]
},
"vendor": {
"type": [
"string",
"null"
]
},
"vram_gb": {
"type": [
"number",
"null"
]
},
"page_url": {
"type": "string",
"description": "Absolute URL of this GPU's live price page, with every live offer."
},
"offer_count": {
"type": "integer"
},
"provider_count": {
"type": "integer"
},
"cheapest_usd_hr": {
"type": [
"number",
"null"
]
},
"cheapest_provider": {
"type": [
"string",
"null"
]
},
"cheapest_min_gpu_count": {
"type": [
"integer",
"null"
]
},
"cheapest_billed_separately": {
"type": [
"string",
"null"
]
}
}
}
},
"count": {
"type": "integer"
},
"stale": {
"type": "boolean"
},
"updated_at": {
"type": [
"string",
"null"
]
}
}
},
"annotations": {
"title": "List cheapest GPU prices",
"readOnlyHint": true,
"openWorldHint": false,
"idempotentHint": true,
"destructiveHint": false
}
}Ranks up to eight GPU configurations that can run a job at the lowest live price. Describe the job in query (an open-weight model to serve or fine-tune, a model size, or a GPU need) or with the structured fields. Each match gives the GPU count, the USD per hour for the whole configuration and a one-line reason; the answer also states the VRAM the job needs and how much cheaper the top match is than a hyperscaler running the same job. Use it to decide where to run a workload; use list_gpu_prices for the price of a named GPU. Open-weight models only: an API-only model such as GPT-4 returns no matches and a note. Ranked on price and fit alone: no provider pays for placement (each match still reports partner, a provider that pays FastGPU a referral). Hard limits (budget_usd_hr, max_total_usd with duration_hours, max_quote_age_minutes) drop every configuration that breaks them; limits reports what was dropped. Read-only: it does not rent, reserve or launch GPUs. No key required.
| Parameter | Type | Description |
|---|---|---|
| spot | string | Set true to include interruptible spot capacity for a cheaper rate. |
| task | string | What the job does. |
| model | string | Open model name to size against, e.g. "Llama 3 70B", "Qwen 72B", "Mixtral". |
| query | string | Plain-language job, e.g. "cheapest to serve Llama 3 70B", "2x H100 for fine-tuning" or "a GPU with 80GB of memory". Provide this OR a structured spec below. |
| region | string | Restrict to a data-residency region. |
| vram_gb | integer | Rough VRAM the job needs, in GB, if you already know it. |
| params_b | number | Model size in billions of parameters when no exact model is named. |
| reserved | string | Set true to include reserved / committed-term capacity for a lower rate. |
| gpu_count | integer | Exact positive GPU count. Overrides a count in query text. Returns no matches if no supported configuration fits; omit for automatic sizing. |
| precision | string | Numeric precision to size the model at. |
| weights_gb | number | Measured size of the weights as stored, in GB: the file you will load (for example a GGUF Q4_K_M build, which holds more bits per parameter than its nominal 4). Replaces parameters times bytes per parameter in the memory estimate. Used for a model sized by its parameter count; the answer's workload.… |
| concurrency | integer | For a serving or fine-tuning job: requests (or sequences) processed at once, 1 to 2048 (default 1). Sizes the KV cache, and for serving the tokens_per_sec and usd_per_million_tokens estimate: more requests at once share each step's weight reads. |
| budget_usd_hr | number | Hard limit on the hourly price of the whole configuration, in USD. Configurations above it are not returned. |
| max_total_usd | number | Hard limit on estimated_total_usd, the GPU rental cost of the whole job, in USD. Needs duration_hours. Configurations above it are not returned. |
| signup_credit | string | Set true to also get new_account_credit_match: the lowest-priced configuration that can run this job at a provider that gives new accounts a credit (confirmed on the provider's own page within 48 hours; FastGPU earns a referral there). It is returned beside matches and never changes them. |
| context_tokens | integer | Tokens of context each request holds, prompt and output together, 512 to 1000000 (default 4096). Sizes the KV cache and the per-token estimate. |
| duration_hours | number | How long the job will run, in hours (fractions are fine). When sent, each match carries billed_hours and estimated_total_usd. |
| active_params_b | number | For a mixture-of-experts model: the parameters it runs per generated token, in billions (Qwen3-30B-A3B runs about 3). Used for tokens_per_sec and usd_per_million_tokens; the memory estimate still holds every expert. Omit it to use FastGPU's figure for a model it knows, or what the Hugging Face repo'… |
| engram_placement | string | For a model with Engram memory tables (DeepSeek-V4.1): gpu (default) holds them in GPU memory; host keeps them in the machine's RAM, as SGLang's --engram-config '{"placement":"cpu"}' does, so vram_required_gb leaves them out and workload.engram_gb is the RAM the machine needs for them. A token reads… |
| min_gpu_memory_gb | integer | Memory each GPU card must have, in GB (e.g. 80 for 80GB-class cards such as the H100 or A100 80GB). Smaller cards are never returned. On its own it lists every card that size, cheapest first. |
| max_quote_age_minutes | number | Hard limit on how old a price quote may be, in minutes. Configurations whose price was fetched from the provider longer ago are not returned. |
{
"name": "match_workload",
"title": "Match a workload to the cheapest GPU",
"description": "Ranks up to eight GPU configurations that can run a job at the lowest live price. Describe the job in query (an open-weight model to serve or fine-tune, a model size, or a GPU need) or with the structured fields. Each match gives the GPU count, the USD per hour for the whole configuration and a one-line reason; the answer also states the VRAM the job needs and how much cheaper the top match is than a hyperscaler running the same job. Use it to decide where to run a workload; use list_gpu_prices for the price of a named GPU. Open-weight models only: an API-only model such as GPT-4 returns no matches and a note. Ranked on price and fit alone: no provider pays for placement (each match still reports partner, a provider that pays FastGPU a referral). Hard limits (budget_usd_hr, max_total_usd with duration_hours, max_quote_age_minutes) drop every configuration that breaks them; limits reports what was dropped. Read-only: it does not rent, reserve or launch GPUs. No key required.",
"inputSchema": {
"type": "object",
"properties": {
"spot": {
"enum": [
"true",
"false"
],
"type": "string",
"description": "Set true to include interruptible spot capacity for a cheaper rate."
},
"task": {
"enum": [
"inference",
"finetune-lora",
"finetune-full",
"generate",
"transcribe",
"embed"
],
"type": "string",
"description": "What the job does."
},
"model": {
"type": "string",
"description": "Open model name to size against, e.g. \"Llama 3 70B\", \"Qwen 72B\", \"Mixtral\"."
},
"query": {
"type": "string",
"description": "Plain-language job, e.g. \"cheapest to serve Llama 3 70B\", \"2x H100 for fine-tuning\" or \"a GPU with 80GB of memory\". Provide this OR a structured spec below."
},
"region": {
"enum": [
"US",
"EU",
"ASIA"
],
"type": "string",
"description": "Restrict to a data-residency region."
},
"vram_gb": {
"type": "integer",
"description": "Rough VRAM the job needs, in GB, if you already know it."
},
"params_b": {
"type": "number",
"description": "Model size in billions of parameters when no exact model is named."
},
"reserved": {
"enum": [
"true",
"false"
],
"type": "string",
"description": "Set true to include reserved / committed-term capacity for a lower rate."
},
"gpu_count": {
"type": "integer",
"description": "Exact positive GPU count. Overrides a count in query text. Returns no matches if no supported configuration fits; omit for automatic sizing."
},
"precision": {
"enum": [
"fp16",
"int8",
"int4"
],
"type": "string",
"description": "Numeric precision to size the model at."
},
"weights_gb": {
"type": "number",
"description": "Measured size of the weights as stored, in GB: the file you will load (for example a GGUF Q4_K_M build, which holds more bits per parameter than its nominal 4). Replaces parameters times bytes per parameter in the memory estimate. Used for a model sized by its parameter count; the answer's workload.weights_gb is null when it was not (an image or speech model sized as a whole pipeline)."
},
"concurrency": {
"type": "integer",
"description": "For a serving or fine-tuning job: requests (or sequences) processed at once, 1 to 2048 (default 1). Sizes the KV cache, and for serving the tokens_per_sec and usd_per_million_tokens estimate: more requests at once share each step's weight reads."
},
"budget_usd_hr": {
"type": "number",
"description": "Hard limit on the hourly price of the whole configuration, in USD. Configurations above it are not returned."
},
"max_total_usd": {
"type": "number",
"description": "Hard limit on estimated_total_usd, the GPU rental cost of the whole job, in USD. Needs duration_hours. Configurations above it are not returned."
},
"signup_credit": {
"enum": [
"true",
"false"
],
"type": "string",
"description": "Set true to also get new_account_credit_match: the lowest-priced configuration that can run this job at a provider that gives new accounts a credit (confirmed on the provider's own page within 48 hours; FastGPU earns a referral there). It is returned beside matches and never changes them."
},
"context_tokens": {
"type": "integer",
"description": "Tokens of context each request holds, prompt and output together, 512 to 1000000 (default 4096). Sizes the KV cache and the per-token estimate."
},
"duration_hours": {
"type": "number",
"description": "How long the job will run, in hours (fractions are fine). When sent, each match carries billed_hours and estimated_total_usd."
},
"active_params_b": {
"type": "number",
"description": "For a mixture-of-experts model: the parameters it runs per generated token, in billions (Qwen3-30B-A3B runs about 3). Used for tokens_per_sec and usd_per_million_tokens; the memory estimate still holds every expert. Omit it to use FastGPU's figure for a model it knows, or what the Hugging Face repo's own config.json states (workload.active_params_source says which)."
},
"engram_placement": {
"enum": [
"gpu",
"host"
],
"type": "string",
"description": "For a model with Engram memory tables (DeepSeek-V4.1): gpu (default) holds them in GPU memory; host keeps them in the machine's RAM, as SGLang's --engram-config '{\"placement\":\"cpu\"}' does, so vram_required_gb leaves them out and workload.engram_gb is the RAM the machine needs for them. A token reads only a few of their rows, so the per-token estimate is the same either way."
},
"min_gpu_memory_gb": {
"type": "integer",
"description": "Memory each GPU card must have, in GB (e.g. 80 for 80GB-class cards such as the H100 or A100 80GB). Smaller cards are never returned. On its own it lists every card that size, cheapest first."
},
"max_quote_age_minutes": {
"type": "number",
"description": "Hard limit on how old a price quote may be, in minutes. Configurations whose price was fetched from the provider longer ago are not returned."
}
}
},
"outputSchema": {
"type": "object",
"required": [
"count",
"matches"
],
"properties": {
"hero": {
"type": [
"object",
"null"
],
"properties": {
"same_model": {
"type": [
"boolean",
"null"
]
},
"savings_pct": {
"type": [
"number",
"null"
]
},
"hyperscaler_ceiling": {
"type": [
"string",
"null"
]
}
}
},
"note": {
"type": "string",
"description": "Why there are no matches, when there are none."
},
"as_of": {
"type": "string",
"description": "When this answer was computed (ISO 8601). Quote ages are measured against it."
},
"count": {
"type": "integer"
},
"error": {
"type": "string",
"description": "Present when the request was not valid; says what to send instead."
},
"stale": {
"type": "boolean"
},
"limits": {
"type": "object",
"properties": {
"excluded": {
"type": "object",
"properties": {
"over_budget_usd_hr": {
"type": "integer"
},
"over_max_total_usd": {
"type": "integer"
},
"older_than_max_quote_age_minutes": {
"type": "integer"
}
}
},
"budget_usd_hr": {
"type": [
"number",
"null"
]
},
"max_total_usd": {
"type": [
"number",
"null"
]
},
"duration_hours": {
"type": [
"number",
"null"
]
},
"nearest_excluded": {
"type": [
"object",
"null"
],
"properties": {
"gpu": {
"type": "string"
},
"url": {
"type": "string",
"description": "Site-relative path of this GPU's live price page, e.g. /gpus/h100-sxm."
},
"score": {
"type": [
"number",
"null"
]
},
"breaks": {
"type": "array",
"items": {
"type": "string"
}
},
"reason": {
"type": "string"
},
"partner": {
"type": "boolean"
},
"page_url": {
"type": "string",
"description": "Absolute URL of this GPU's live price page, with every live offer."
},
"provider": {
"type": "string"
},
"rent_url": {
"type": "string",
"description": "Opens this offer on the provider's own site, through a FastGPU redirect that counts the click. When partner is true the link carries FastGPU's referral code (the provider pays FastGPU a referral; the price is the same). Give it to the user only after they choose this offer."
},
"gpu_count": {
"type": "integer"
},
"fetched_at": {
"type": [
"string",
"null"
],
"description": "When this price was fetched from the provider (ISO 8601)."
},
"offer_type": {
"type": "string"
},
"monthly_usd": {
"type": "number"
},
"over_budget": {
"type": "boolean"
},
"reliability": {
"type": "string"
},
"token_model": {
"type": [
"object",
"null"
],
"properties": {
"bound": {
"enum": [
"memory",
"compute",
"latency"
],
"type": "string"
},
"requests": {
"type": "integer"
},
"min_step_ms": {
"type": "number"
},
"cards_counted": {
"type": "integer"
},
"kv_gb_per_step": {
"type": "number"
},
"bytes_per_param": {
"type": "number"
},
"memory_efficiency": {
"type": "number"
},
"streamed_params_b": {
"type": "number"
},
"card_bandwidth_gbs": {
"type": "number"
},
"weights_gb_per_step": {
"type": "number"
},
"memory_tokens_per_sec": {
"type": "number"
},
"compute_tokens_per_sec": {
"type": "number"
},
"latency_tokens_per_sec": {
"type": "number"
}
},
"description": "The working behind tokens_per_sec (serving only), a model and not a measurement: each decode step reads streamed_params_b x bytes_per_param of weights (weights_gb_per_step) plus every request's KV cache (kv_gb_per_step); tokens_per_sec is the lowest of memory_tokens_per_sec = requests x cards_counted x card_bandwidth_gbs x memory_efficiency / (weights_gb_per_step + kv_gb_per_step), the compute ceiling compute_tokens_per_sec, and latency_tokens_per_sec = requests / min_step_ms (no decode step is credited faster than the fastest published single-request decode). A best case: real serving stacks usually reach less."
},
"billed_hours": {
"type": [
"number",
"null"
],
"description": "Hours the provider bills for duration_hours: rounded up to whole hours on a per-hour meter and to whole minutes on a per-minute meter. Null when duration_hours was not sent."
},
"provider_label": {
"type": [
"string",
"null"
]
},
"tokens_per_sec": {
"type": [
"number",
"null"
]
},
"effective_usd_hr": {
"type": "number"
},
"fits_single_card": {
"type": "boolean"
},
"billing_increment": {
"enum": [
"per-second",
"per-minute",
"per-hour",
"varies"
],
"type": "string",
"description": "How this provider's meter ticks for on-demand rental; varies means FastGPU has no verified figure."
},
"quote_age_seconds": {
"type": [
"integer",
"null"
],
"description": "Seconds between fetched_at and as_of. Null when the fetch time is unknown."
},
"new_account_credit": {
"type": [
"object",
"null"
],
"properties": {
"usd": {
"type": "number"
},
"terms_url": {
"type": "string"
},
"confirmed_at": {
"type": "string"
},
"expires_days": {
"type": [
"integer",
"null"
]
}
},
"description": "A credit the provider gives a NEW account opened through rent_url, shown only while FastGPU read it on the provider's own referral page within the last 48 hours; null when none is confirmed. usd is the credit, expires_days how long its unused part lasts (null when the provider states none), confirmed_at when FastGPU last read it there, terms_url the FastGPU page that quotes the provider's conditions and what FastGPU earns. It never changes the ranking or the price."
},
"estimated_total_usd": {
"type": [
"number",
"null"
],
"description": "effective_usd_hr times billed_hours, rounded up to the cent. GPU rental only: storage, data transfer, CPU and memory billed separately, and provider minimums are not included. Null when duration_hours was not sent."
},
"usd_per_million_tokens": {
"type": [
"number",
"null"
]
}
},
"description": "When the limits leave no match: the cheapest configuration that can still run the job, and the limits it breaks."
},
"max_quote_age_minutes": {
"type": [
"number",
"null"
]
}
},
"description": "The hard limits this answer applied (null when one was not sent) and how many configurations each one left out."
},
"matches": {
"type": "array",
"items": {
"type": "object",
"properties": {
"gpu": {
"type": "string"
},
"url": {
"type": "string",
"description": "Site-relative path of this GPU's live price page, e.g. /gpus/h100-sxm."
},
"score": {
"type": [
"number",
"null"
]
},
"reason": {
"type": "string"
},
"partner": {
"type": "boolean"
},
"page_url": {
"type": "string",
"description": "Absolute URL of this GPU's live price page, with every live offer."
},
"provider": {
"type": "string"
},
"rent_url": {
"type": "string",
"description": "Opens this offer on the provider's own site, through a FastGPU redirect that counts the click. When partner is true the link carries FastGPU's referral code (the provider pays FastGPU a referral; the price is the same). Give it to the user only after they choose this offer."
},
"gpu_count": {
"type": "integer"
},
"fetched_at": {
"type": [
"string",
"null"
],
"description": "When this price was fetched from the provider (ISO 8601)."
},
"offer_type": {
"type": "string"
},
"monthly_usd": {
"type": "number"
},
"over_budget": {
"type": "boolean"
},
"reliability": {
"type": "string"
},
"token_model": {
"type": [
"object",
"null"
],
"properties": {
"bound": {
"enum": [
"memory",
"compute",
"latency"
],
"type": "string"
},
"requests": {
"type": "integer"
},
"min_step_ms": {
"type": "number"
},
"cards_counted": {
"type": "integer"
},
"kv_gb_per_step": {
"type": "number"
},
"bytes_per_param": {
"type": "number"
},
"memory_efficiency": {
"type": "number"
},
"streamed_params_b": {
"type": "number"
},
"card_bandwidth_gbs": {
"type": "number"
},
"weights_gb_per_step": {
"type": "number"
},
"memory_tokens_per_sec": {
"type": "number"
},
"compute_tokens_per_sec": {
"type": "number"
},
"latency_tokens_per_sec": {
"type": "number"
}
},
"description": "The working behind tokens_per_sec (serving only), a model and not a measurement: each decode step reads streamed_params_b x bytes_per_param of weights (weights_gb_per_step) plus every request's KV cache (kv_gb_per_step); tokens_per_sec is the lowest of memory_tokens_per_sec = requests x cards_counted x card_bandwidth_gbs x memory_efficiency / (weights_gb_per_step + kv_gb_per_step), the compute ceiling compute_tokens_per_sec, and latency_tokens_per_sec = requests / min_step_ms (no decode step is credited faster than the fastest published single-request decode). A best case: real serving stacks usually reach less."
},
"billed_hours": {
"type": [
"number",
"null"
],
"description": "Hours the provider bills for duration_hours: rounded up to whole hours on a per-hour meter and to whole minutes on a per-minute meter. Null when duration_hours was not sent."
},
"provider_label": {
"type": [
"string",
"null"
]
},
"tokens_per_sec": {
"type": [
"number",
"null"
]
},
"effective_usd_hr": {
"type": "number"
},
"fits_single_card": {
"type": "boolean"
},
"billing_increment": {
"enum": [
"per-second",
"per-minute",
"per-hour",
"varies"
],
"type": "string",
"description": "How this provider's meter ticks for on-demand rental; varies means FastGPU has no verified figure."
},
"quote_age_seconds": {
"type": [
"integer",
"null"
],
"description": "Seconds between fetched_at and as_of. Null when the fetch time is unknown."
},
"new_account_credit": {
"type": [
"object",
"null"
],
"properties": {
"usd": {
"type": "number"
},
"terms_url": {
"type": "string"
},
"confirmed_at": {
"type": "string"
},
"expires_days": {
"type": [
"integer",
"null"
]
}
},
"description": "A credit the provider gives a NEW account opened through rent_url, shown only while FastGPU read it on the provider's own referral page within the last 48 hours; null when none is confirmed. usd is the credit, expires_days how long its unused part lasts (null when the provider states none), confirmed_at when FastGPU last read it there, terms_url the FastGPU page that quotes the provider's conditions and what FastGPU earns. It never changes the ranking or the price."
},
"estimated_total_usd": {
"type": [
"number",
"null"
],
"description": "effective_usd_hr times billed_hours, rounded up to the cent. GPU rental only: storage, data transfer, CPU and memory billed separately, and provider minimums are not included. Null when duration_hours was not sent."
},
"usd_per_million_tokens": {
"type": [
"number",
"null"
]
}
}
}
},
"workload": {
"type": "object",
"properties": {
"task": {
"type": "string"
},
"experts": {
"type": [
"integer",
"null"
]
},
"engram_gb": {
"type": [
"number",
"null"
],
"description": "The Engram tables' size at this sizing, in GB: in vram_required_gb when engram_placement is gpu, the host RAM they need when it is host."
},
"gpu_count": {
"type": [
"integer",
"null"
]
},
"precision": {
"type": "string"
},
"identified": {
"type": "boolean"
},
"concurrency": {
"type": "integer",
"description": "Requests the job is sized for at once."
},
"gpu_vendors": {
"type": [
"array",
"null"
],
"items": {
"type": "string"
}
},
"context_tokens": {
"type": "integer",
"description": "Context each request holds, in tokens."
},
"experts_active": {
"type": [
"integer",
"null"
]
},
"interpretation": {
"type": "string"
},
"active_params_b": {
"type": [
"number",
"null"
],
"description": "Parameters a mixture-of-experts model runs per generated token, in billions; null for a dense model."
},
"engram_params_b": {
"type": [
"number",
"null"
],
"description": "Parameters held in Engram lookup tables, in billions (DeepSeek-V4.1); null when the model has none."
},
"engram_placement": {
"enum": [
"gpu",
"host",
null
],
"type": [
"string",
"null"
]
},
"vram_required_gb": {
"type": "number"
},
"min_gpu_memory_gb": {
"type": [
"integer",
"null"
]
},
"active_params_source": {
"enum": [
"stated",
"registry",
"name",
"config",
null
],
"type": [
"string",
"null"
],
"description": "Where active_params_b comes from: the request, FastGPU's model list, the repo's name, or the repo's config.json."
}
}
},
"updated_at": {
"type": [
"string",
"null"
]
},
"new_account_credit_match": {
"type": [
"object",
"null"
],
"properties": {
"gpu": {
"type": "string"
},
"url": {
"type": "string",
"description": "Site-relative path of this GPU's live price page, e.g. /gpus/h100-sxm."
},
"score": {
"type": [
"number",
"null"
]
},
"reason": {
"type": "string"
},
"partner": {
"type": "boolean"
},
"page_url": {
"type": "string",
"description": "Absolute URL of this GPU's live price page, with every live offer."
},
"provider": {
"type": "string"
},
"rent_url": {
"type": "string",
"description": "Opens this offer on the provider's own site, through a FastGPU redirect that counts the click. When partner is true the link carries FastGPU's referral code (the provider pays FastGPU a referral; the price is the same). Give it to the user only after they choose this offer."
},
"gpu_count": {
"type": "integer"
},
"fetched_at": {
"type": [
"string",
"null"
],
"description": "When this price was fetched from the provider (ISO 8601)."
},
"offer_type": {
"type": "string"
},
"monthly_usd": {
"type": "number"
},
"over_budget": {
"type": "boolean"
},
"reliability": {
"type": "string"
},
"token_model": {
"type": [
"object",
"null"
],
"properties": {
"bound": {
"enum": [
"memory",
"compute",
"latency"
],
"type": "string"
},
"requests": {
"type": "integer"
},
"min_step_ms": {
"type": "number"
},
"cards_counted": {
"type": "integer"
},
"kv_gb_per_step": {
"type": "number"
},
"bytes_per_param": {
"type": "number"
},
"memory_efficiency": {
"type": "number"
},
"streamed_params_b": {
"type": "number"
},
"card_bandwidth_gbs": {
"type": "number"
},
"weights_gb_per_step": {
"type": "number"
},
"memory_tokens_per_sec": {
"type": "number"
},
"compute_tokens_per_sec": {
"type": "number"
},
"latency_tokens_per_sec": {
"type": "number"
}
},
"description": "The working behind tokens_per_sec (serving only), a model and not a measurement: each decode step reads streamed_params_b x bytes_per_param of weights (weights_gb_per_step) plus every request's KV cache (kv_gb_per_step); tokens_per_sec is the lowest of memory_tokens_per_sec = requests x cards_counted x card_bandwidth_gbs x memory_efficiency / (weights_gb_per_step + kv_gb_per_step), the compute ceiling compute_tokens_per_sec, and latency_tokens_per_sec = requests / min_step_ms (no decode step is credited faster than the fastest published single-request decode). A best case: real serving stacks usually reach less."
},
"billed_hours": {
"type": [
"number",
"null"
],
"description": "Hours the provider bills for duration_hours: rounded up to whole hours on a per-hour meter and to whole minutes on a per-minute meter. Null when duration_hours was not sent."
},
"provider_label": {
"type": [
"string",
"null"
]
},
"tokens_per_sec": {
"type": [
"number",
"null"
]
},
"effective_usd_hr": {
"type": "number"
},
"fits_single_card": {
"type": "boolean"
},
"billing_increment": {
"enum": [
"per-second",
"per-minute",
"per-hour",
"varies"
],
"type": "string",
"description": "How this provider's meter ticks for on-demand rental; varies means FastGPU has no verified figure."
},
"quote_age_seconds": {
"type": [
"integer",
"null"
],
"description": "Seconds between fetched_at and as_of. Null when the fetch time is unknown."
},
"new_account_credit": {
"type": [
"object",
"null"
],
"properties": {
"usd": {
"type": "number"
},
"terms_url": {
"type": "string"
},
"confirmed_at": {
"type": "string"
},
"expires_days": {
"type": [
"integer",
"null"
]
}
},
"description": "A credit the provider gives a NEW account opened through rent_url, shown only while FastGPU read it on the provider's own referral page within the last 48 hours; null when none is confirmed. usd is the credit, expires_days how long its unused part lasts (null when the provider states none), confirmed_at when FastGPU last read it there, terms_url the FastGPU page that quotes the provider's conditions and what FastGPU earns. It never changes the ranking or the price."
},
"estimated_total_usd": {
"type": [
"number",
"null"
],
"description": "effective_usd_hr times billed_hours, rounded up to the cent. GPU rental only: storage, data transfer, CPU and memory billed separately, and provider minimums are not included. Null when duration_hours was not sent."
},
"usd_per_million_tokens": {
"type": [
"number",
"null"
]
}
},
"description": "Only when signup_credit=true was sent: the lowest-priced configuration that can run this job at a provider giving new accounts a confirmed credit (its new_account_credit says how much), wherever it ranks; null when no such provider can run it. FastGPU earns a referral from that provider. It is beside matches and never changes them."
}
}
},
"annotations": {
"title": "Match a workload to the cheapest GPU",
"readOnlyHint": true,
"openWorldHint": false,
"idempotentHint": true,
"destructiveHint": false
}
}Compare live GPU cloud rental prices across the whole market and route a workload to the cheapest fit. Available tools: list_gpu_prices, match_workload.
Names, descriptions and schemas were written by the server and are shown as data.
Shared addresses on big platforms (Cloudflare, Vercel and the like) put unrelated servers together; those are marked shared_infrastructure. A match is a pivot, not a link between operators.