Public prices for fifty agent platforms.

Frontier model API. Compared on per-token, cached-input, and batch prices.

“Claude Opus 5.5 | $20 / MTok output”

Frontier model API. Standard and batch token prices from the API pricing page.

“gpt-6-astra | Short context output | $50.00”

Gemini Developer API. Standard paid rates, including prices that the page says change on a stated date.

“Output price (including thinking tokens) | $10.00 through November 30, 2026. $20.00 starting December 1, 2026.”

Grok API. Short-context rates below 200k prompt tokens, plus the published rate limits.

“Output | < 200k prompt tokens | $6.00”

API prices for Mistral and Ministral models, plus open-weight flags from the model overview.

“Mistral Large 4 | Output | $4.18”

Per-million-token prices from the API docs, split into the published peak and off-peak rates.

“deepseek-flash | 1M OUTPUT TOKENS | OFF-PEAK | $0.6”

api-docs.deepseek.com/quick_start/pricing and deepseek.com/en/platform listed different USD figures on the same check. Rows use the API docs table. The marketing page was not used as a price.

Enterprise model API. Legacy Command token rates are on the pricing FAQ. Current token rates were not listed.

“Command pricing is $1.00/1M tokens for input and $2.00/1M tokens for output”

The pricing page states legacy Command token rates for existing customers. Model Vault hourly rates rendered without numbers. Current Command A token prices are left empty.

Multi-model API on AWS. Only rows whose prices rendered on the public pricing page are filled in.

“DeepSeek v3.2 | $ 0.62 | $ 1.85”

The Anthropic on-demand table on the Bedrock pricing page did not render in this fetch. Provisioned throughput says to contact an account team. Those prices are empty. Rows that did render are stored.

Azure OpenAI and model catalog. No list price was extracted.

The Azure OpenAI pricing URL timed out on fetch. No price was stored.

Agent Platform generative pricing. Global endpoint rates from the first Gemini 4 Argon and Gemini 3.8 Flash tables.

“Text output (response and reasoning) | Global | $10”

Later tables on the same page list different rates. Their tier labels did not survive extraction, so only the first global table for each model is stored.

OCI Generative AI. The public pricing URL resolved to a cost estimator without a list-price table.

Fetched the pricing URL. The page is an OCI Enterprise AI cost estimator and did not publish a token or GPU list price in the extract. Prices are empty.

Serverless open-model prices and per-GPU-hour rates for dedicated inference and GPU clusters.

“DeepSeek V4.1 Flash | $1.20”

The serverless table was printed under a 'Batch API price' label as well as 'Price per 1M tokens'. The toggle state did not survive extraction. Token rows are the numbers in that table, and the tier is marked ambiguous in the row note.

Serverless per-token rates and on-demand GPU-hour rates.

“OpenAI GPT OSS 120B | output | $0.60”

Dedicated model serving. Compliance is published. GPU and token units on the pricing page were not reliable enough to store.

The pricing page rendered model and GPU figures without a stable unit label next to each number. Those figures were not stored. SOC 2 Type II and HIPAA were stated in the FAQ.

LPU inference. Token prices and tokens/s from the GroqDocs models page. Llama rows there say Contact Sales.

“GPT OSS 120B openai/gpt-oss-120b | $0.60 output”

home.cloud.groq.io/pricing listed dollar prices for Llama 3.1 8B and Llama 3.3 70B. console.groq.com/docs/models lists those production models as Contact Sales. The docs page is the source used here, so those Llama prices are empty.

Wafer-scale inference. No confirmed live price was stored.

The public models docs include a sample JSON body with pricing for gpt-oss-120b. The unauthenticated models endpoint returned HTTP 403, so the sample was not confirmed and was not stored as a price. cerebras.ai/pricing did not print per-token numbers in the fetch.

RDU inference. Per-million-token rates from the SambaCloud pricing table.

“gpt-oss-120b | Output (per 1M tokens) | 0.59 USD”

Serverless per-token rates and dedicated GPU-hour rates.

“DeepSeek-V4-Flash | $0.18”

Router across inference providers. The pricing page describes pass-through billing, not a model price list.

No per-model token table was on the pricing doc. Model prices are left empty rather than copied from routed providers.

Serverless GPU billed per second, scale to zero. SOC 2 is included on the published plans.

“Nvidia H100 SXM5 $0.001097 / sec”

GPU cloud. Prices below are the published on-demand node prices, not a per-GPU rate.

“NVIDIA HGX H100 | GPU Count 8 | On-Demand Price: $49.24 / Hour”

HGX H100 is listed as GPU count 8 at $49.24/hour on-demand. That node price is stored as published. It is not divided into a per-GPU rate, and it is excluded from the run-cost model.

GPU cloud. On-demand GPU-hour rates. Spot is described as dynamic, so no spot number is stored.

“NVIDIA HGX H100 | On-demand, GPU-hour | $4.50”

The prices page says spot is 'from' a figure and changes in the console. Spot cells are empty. Region names were not on this page.

On-demand GPU instances. Rates are per GPU-hour and depend on the instance width.

“Instances pricing, 8× | NVIDIA H100 SXM | 80 GB | PRICE/GPU/HR | $3.99”

GPU instances and self-serve dedicated deployments, priced per GPU-hour.

“NVIDIA H100 | 80GB HGX | $4.50/GPU-hr”

GPU pods priced per hour on the public pricing page.

“H100 SXM | 80 GB VRAM | $3.99/hr”

Secure Cloud pod rates that rendered next to a named GPU are stored. A marketing line 'from $1.99/hr' was not used as a SKU price.

Open serving engine. No inference price and no single tokens/s figure was fetched.

No pricing page. Throughput depends on the GPU, model, and build. Tokens/s is empty. Enter a measured rate in Compare to price a GPU SKU.

Open serving engine aimed at prefix-heavy workloads. No price and no single tokens/s figure was fetched.

No pricing page. Tokens/s is empty.

NVIDIA serving stack. No public tokens/s or software price was fetched for this row.

No single public price or tokens/s number was fetched. Left empty.

Local serving for laptop and edge. No cloud price.

Local runtime. No cloud token or GPU-hour price.

Managed gateway. A few model rates from the public models API, fetched with the catalog.

“anthropic/claude-sonnet-5.5 pricing.completion 0.00001 USD per token ($10 / 1M).”

The models API lists hundreds of routes. Three rows are stored as a sample of that feed, not a full mirror.

Open-source gateway. No hosted token price was fetched.

No hosted inference price was fetched. Price fields are empty.

Gateway and observability. The production plan price is a platform fee, not a model token price.

Enterprise compliance (SOC2 Type 2, GDPR, HIPAA) is listed on the custom enterprise plan, not on the $49 production plan, so it is not treated as a default certification.

Edge gateway. A model price list was not extracted.

The docs index was fetched (HTTP 200) and did not yield a token price table in this pass. Prices are empty.

Provider routing for TypeScript. Token charges are the provider list price, with published add-on meters.

Agent traces. A public seat or trace price was not extracted in this pass.

langchain.com/pricing returned HTTP 200. A trace or seat price was not extracted. Left empty.

Open-source LLM observability. A plan price was not extracted.

langfuse.com/pricing returned HTTP 200. A plan price was not extracted. Left empty.

Evals and observability. A plan price was not extracted.

braintrust.dev/pricing returned HTTP 200. A plan price was not extracted. Left empty.

Open-source Phoenix and commercial AX. A plan price was not extracted.

arize.com/pricing returned HTTP 200. A plan price was not extracted. Left empty.

APM extension for LLM spans. A public LLM-observability unit price was not extracted.

datadoghq.com/pricing returned HTTP 200. An LLM Observability unit price was not extracted. Left empty.

Trace standard, not a priced product.

A specification. No price.

Agent framework. Graph nodes and model calls. No inference price.

Framework. No inference price was fetched.

Agent framework. Agents, handoffs, and tools. Model calls are billed by the API you point it at.

Framework. No separate inference price.

Agent framework for the Claude agent loop. Model calls are billed on the Claude API.

Framework. Token prices live on the Claude API row.

Agent framework. Crews, roles, and task sequencing. No inference price.

Framework. No inference price was fetched.

Agent and RAG framework. No inference price.

Framework. No inference price was fetched.

Typed agent framework. No inference price.

Framework. No inference price was fetched.

Agent framework for Gemini and Vertex. No separate inference price.

Framework. Token prices live on the Gemini and Vertex rows.

Agent framework. Successor to Semantic Kernel and AutoGen. No inference price.

Framework. No inference price was fetched.

TypeScript agent framework. No inference price.

Framework. No inference price was fetched.

TypeScript SDK for generateText and streamText. Model calls are billed by the provider.

SDK. No inference price of its own. Gateway add-on meters are on the Vercel AI Gateway row.