21 tools compared on real pricing, genuine strengths, and the limitations that only show up after you buy.
We track 21 AI Infrastructure tools, 13 of which have a free or freemium tier you can evaluate without a sales call. Our picks are Amazon Bedrock, Browser Use and Databricks Mosaic AI. Each entry below lists real pricing, what it is genuinely good at, and the limitations that only show up after you buy.
| Tool | Pricing model | What it costs |
|---|---|---|
| Amazon Bedrock | paid | Usage-based pricing (per token, or provisioned throughput). Billed through AWS. |
| Browser Use | freemium | The Python library is open source and free to self-host. Cloud Browser Infrastructure is $0.02 per browser-hour, metered by the minute. Cloud Agents bill browser time from $0.02/hour plus model tokens at published per-million rates (their default model lists at $0.24 per million input tokens and $1.44 per million output). Usage-based throughout — no seat fee. |
| Databricks Mosaic AI | enterprise-quoted | Databricks now sells its agent tooling as Agent Bricks, billed in DBUs; the rates below are for Premium on AWS US East. Model Serving CPU costs $0.07/DBU, with standard workloads billed at 4 DBU per hour. Databricks Apps, which Agent Runtime runs on, bill App Capacity at $0.75/DBU, and the default Medium app size uses 0.5 DBU per hour. Databricks publishes no separate Agent Runtime price, so confirm whether it is billed as App Capacity. Proprietary-model pay-per-token costs $0.07/DBU. The Knowledge Assistant ($0.15 per answer) and Supervisor Agent are no longer available to new customers. As read on the vendor's pricing page on 10 November 2026. |
| Amazon Bedrock AgentCore | enterprise-quoted | Consumption-based, with each AgentCore component metered separately. Runtime v2 costs $0.1276 per vCPU-hour (CPU billed only while active) and $0.0169 per GB-hour, billed per second, with idle memory reclaimed after 120 seconds. Gateway costs $0.005 per 1,000 invocations. Short-term memory costs $1.00 per GB ingested (12 KB minimum per event), and long-term memory costs $0.75 per 1,000 records/month plus $0.50 per 1,000 retrievals. Identity is free through Runtime or Gateway. Browser and Code Interpreter cost $0.0895 per vCPU-hour, and Observability is billed at CloudWatch rates. Model inference is billed separately. As read on the vendor's pricing page on 10 November 2026. |
| Arcade.dev | freemium | Free is $0 a month with 2,000 tool calls; Team is a $25 monthly platform fee plus $0.01 per tool call with next-business-day email support; Enterprise is custom with annual tool-call bundles, discounted additional call rates, and deployment on Arcade Cloud, in your VPC or fully air-gapped. Priced by what agents do, not per seat. As read on the vendor's pricing page on 10 November 2026. |
| Azure OpenAI Service | paid | Usage-based pricing, billed through Azure. Provisioned throughput units available for guaranteed capacity. |
| Baseten | paid | Usage-based pricing tied to the compute your deployed models consume. |
| Composio | freemium | Developer platform: Hobby is $0 with 100,000 tool calls, 50,000 trigger events and 3 team members a month, hard-capped (20,000 of those calls on Composio-managed OAuth apps); Pro is $29 a month including $29 of usage credit, then $0.0003 per tool call and $0.003 per trigger event, with add-ons such as a HIPAA BAA at +$0.0003 per call; Enterprise is custom with SSO/SCIM and customer-managed keys (Composio's homepage also lists own-VPC and fully self-hosted deployment). These developer-platform plans apply to signups on or after 15 August 2026; earlier customers stay on their existing plan through 31 December 2026. Separate "For You" team plans are $18 per seat per month. As read on the vendor's pricing page on 10 November 2026. |
| E2B | freemium | Hobby is free with $100 one-time usage credits, community support, sessions up to 1 hour, and 20 concurrent sandboxes; no credit card required. Pro is $150/month with customisable sandbox CPU and RAM, sessions up to 24 hours, and 100 concurrent sandboxes, expandable by purchase to 1,100. Enterprise is custom. Compute is billed per second on vCPU, RAM, and storage on top of the plan fee. |
| Exa | freemium | Free trial via the dashboard; usage-based pricing published on the pricing page rather than the homepage. |
| Firecrawl | freemium | Free 1,000 credits/month. Hobby $16/mo for 5,000. Standard $83/mo for 100,000 with 50 concurrent requests. Growth $333/mo for 500,000. Scale $599/mo for 1,000,000. Enterprise custom. Prices billed yearly. Scrape, crawl, map, and monitor cost 1 credit per page; search costs 2 credits per 10 results. |
| Gemini Enterprise Agent Platform (formerly Vertex AI) | paid | Vertex AI is now Gemini Enterprise Agent Platform, billed by usage through Google Cloud. Agent Runtime (formerly Agent Engine) bills allocated resources: Agent Compute at $0.085 per vCPU-hour and Agent Memory at $0.009 per GiB-hour, with idle time between turns not billed. Agent Storage costs $0.30 per GiB-month. Memory Bank and Sessions are billed at 1 vCPU-hour per 1 million writes or 3 million reads. 1- and 3-year Flexible Savings Plans lower compute to $0.0765 and $0.068. Model usage is billed per model. As read on the vendor's pricing page on 10 November 2026. |
| Groq | freemium | Usage-based, per million tokens. The self-serve production list is short: GPT OSS 120B $0.15 input / $0.60 output, GPT OSS 20B $0.075 / $0.30, plus Whisper speech-to-text. Llama models are listed under Enterprise, "Contact Sales". As read on the vendor's site on 10 November 2026. |
| LiteLLM | freemium | Open-source and free to self-host. A paid enterprise edition adds SSO, audit logs, and support. |
| Mem0 | freemium | Open source and free to self-host. A managed platform with a free starting tier is available; paid tier pricing is on the pricing page rather than the homepage. |
| Modal | freemium | Usage-based compute pricing with a recurring free credit allowance for getting started. |
| OpenRouter | paid | Model prices are the provider's own, passed through without markup. Free plan: free models only, 50 requests a day. Pay as you go: a 5.5% fee when you buy credits (8% on Business, which adds EU and US in-region routing). Bring your own provider keys and the first $25,000 a month of list-price usage carries no fee, then 5%. As read on the vendor's pricing page and FAQ on 10 November 2026. |
| Prisma AIRS AI Gateway (formerly Portkey) | freemium | Free developer tier. Paid Pro and Enterprise plans add volume, self-hosting, and governance. |
| Temporal | freemium | Open-source and self-hostable for free. Temporal Cloud is a usage-based managed service. |
| Together AI | paid | Usage-based per-token pricing. Dedicated endpoints and fine-tuning are priced separately. |
| Unstructured | freemium | Open-source libraries are free. The managed API and platform are sold on usage-based paid plans. |
AWS's fully managed service for accessing foundation models from multiple providers, with agents, guardrails, and knowledge bases built in.
Give an agent a natural-language task and it drives a real browser to completion — open source library, or managed cloud browsers.
Databricks' suite for building, governing, and serving production AI — agents, RAG, fine-tuning, and evaluation — on top of your governed data.
AWS framework-agnostic runtime for deploying, securing, evaluating, and governing production agents at scale.
One control point for the actions agents take in real systems — user-scoped auth, 8,000+ permission-aware tools, and an audit trail per action.
Microsoft Azure's managed access to OpenAI models, deployed within an enterprise Azure tenant with its security and compliance controls.
A platform for deploying and serving machine-learning models in production, with autoscaling, fast cold starts, and GPU infrastructure managed for you.
Gives agents authenticated access to 1,000+ applications, handling OAuth, tool selection, and sandboxed execution so you do not build integrations one at a time.
Isolated sandboxes where an AI agent can run the code it just wrote without touching anything you care about.
A search API built for AI agents rather than humans — structured outputs, token-efficient highlights, and sub-200ms responses.
Turns any website into clean, structured markdown or JSON for AI pipelines — scrape, crawl, map, and search with published per-credit pricing.
Google Cloud's unified AI platform — access to Gemini and partner models, plus tools to build, deploy, and govern AI and agents.
An inference provider whose custom LPU hardware delivers exceptionally low-latency responses for open-weight models.
An open-source LLM gateway that gives you one consistent API and proxy across 100+ model providers, with key management and spend tracking.
A memory layer for AI agents that condenses conversation history into compact, retrievable memories — cutting tokens and latency while keeping context.
A serverless cloud for running AI and data workloads — define infrastructure in Python and get on-demand GPUs without managing servers.
A unified API and marketplace that routes requests to hundreds of models from many providers through a single endpoint and bill.
An AI gateway that adds routing, caching, observability, and guardrails to LLM traffic through a single control plane.
A durable execution platform that makes long-running, failure-prone workflows — including AI agents — reliable by default.
A cloud platform for fast, cost-efficient inference and fine-tuning of open-weight models at production scale.
A platform for turning messy enterprise documents — PDFs, slides, emails, scans — into clean, structured data ready for RAG and LLMs.
Practical playbooks, prompts, and tool stacks — for the SMB operators and enterprise teams actually shipping AI.
Free implementation guides for SMB operators and enterprise teams — workflows, prompts, governance, and tool stacks built to ship.