Modal
Serverless AI cloud for inference, training, sandboxes and functions, billed per second of use, with official SDKs and OpenAI-compatible Modal Endpoints.
Tags
Description
Vendor-affiliated listing (Modal Labs). Established platform, not a claim of a brand-new launch. https://www.modal.com 301s to https://modal.com/; use the apex host. Docs https://modal.com/docs (200; https://modal.com/docs/ 308s to /docs and is not used). Official intro and llms.txt: Modal hosts the compute and charges per second of usage; containers scale to zero when idle. The Python SDK/CLI (pip install modal, uv add modal, or conda install conda-forge::modal-client) is the primary interface; JavaScript/TypeScript and Go SDKs can call Functions, run Sandboxes, and manage resources. A standalone CLI install script curl -LsSf uvx.sh/modal/install.sh | sh is also documented and was not executed. Auth: after installing the CLI, modal setup (or python -m modal setup) creates a Modal token in a browser session; modal token info verifies it. Tokens were not created. Pricing (https://modal.com/pricing, 200, and https://modal.com/docs/guide/billing, 200): you pay for compute used or requested, with no required reservations and no minimum usage-time increments. Workspaces are billed monthly and may also be auto-charged mid-cycle when usage first exceeds certain thresholds. A payment method must be on file to use Modal; the intro also says using a GPU requires a valid payment method. The pricing page lists a Starter plan at $0 + compute / month with $30 / month included compute credit, and a Team plan at $250 + compute / month with $100 / month included credit (Enterprise is custom). Official llms.txt likewise says you can sign up and get $30/month of credits. Shared Endpoint docs state that credits included with a plan cannot be used for Shared Endpoint usage. Live unit prices (GPU/CPU/memory/volumes/egress and Model API per-1M-token rates) are on the pricing page and are not copied here. Refund terms were not independently extracted from https://modal.com/legal/terms (200). No signup or spend was used. Modal Endpoints deploy supported or custom models as inference APIs. Shared Endpoints are billed per token; Dedicated Endpoints are billed for compute resources and can scale to zero. Text-generation endpoints can be called with OpenAI-compatible Chat Completions and Responses APIs or the Anthropic-compatible Messages API; embeddings have an OpenAI-compatible Embeddings API. Shared Endpoints always need a Proxy Token (modal workspace proxy-tokens create, passed as Authorization: Bearer wk-<id>.ws-<secret> and usable as the API key in an OpenAI-compatible client). Dedicated Endpoints require a proxy token by default and can also be created --unauthenticated. Docs also include an example of deploying your own OpenAI-compatible LLM service with vLLM. Sandboxes are isolated environments for coding agents, RL rollouts, and untrusted code (code execution / side effects). The official client repo https://github.com/modal-labs/modal-client is Apache-2.0 (not archived); the hosted service is proprietary. pricing freemium (Starter includes monthly credits; a payment method is still required). https://modal.com/llms.txt and https://modal.com/legal/privacy-policy were HTTP 200. https://modal.com/privacy and https://modal.com/legal/privacy returned 404 and are not used.
Agent access
Related
- E2B: Open-source, secure cloud sandboxes for AI agents: an isolated machine per agent to run code, browse and use tools.
Guides & comparisons
- Run agent-generated code safely in a sandbox (Guide): A defense-in-depth recipe for executing LLM-written or user-supplied code: pick an isolation boundary, lock down egress, keep secrets out, cap time and resources, and treat the output as untrusted.
- GPU clouds and serverless inference for agents (Comparison): Compare GPU rental, serverless GPUs and hosted per-token APIs for agents: OpenAI-compatible base URLs, tools and structured output, BYO models, scale-to-zero and billing, APIs/CLIs, and host-trust notes. No cached dollar figures or invented benchmarks.
- Agent environment frameworks compared: Harbor vs AgentEnv Framework vs Seahaven (Comparison): Three open-source Python frameworks for building the environments agent evals and RL runs happen in: Harbor (container tasks), Scale AI's AgentEnv Framework and Kiln AI's Seahaven (synthetic SQLite worlds). Covers tasks, grading, compute, agent interface, interop and license.
- Agent benchmarks compared (Comparison): Ten widely cited agent benchmarks (SWE-bench, Terminal-Bench, tau-bench, BFCL, MCPMark, GAIA, BrowseComp, WebArena, Mind2Web and OSWorld) compared on what they measure, how many tasks they have, where the agent runs, how runs are scored, where results are published and how they are licensed.
- Code sandboxes for AI agents (Comparison): E2B, Daytona, Modal, Vercel Sandbox, Cloudflare Sandboxes, Fly.io Sprites and microsandbox compared on isolation, pricing, free tier, lifetime, persistence, egress controls, SDKs, GPUs and self-hosting.
Sources
- https://modal.com
- https://modal.com/pricing
- https://modal.com/docs
- https://modal.com/llms.txt
- https://modal.com/docs/guide/intro
- https://modal.com/docs/guide/getting-started
- https://modal.com/docs/guide/billing
- https://modal.com/docs/guide/endpoints
- https://modal.com/docs/guide/shared-endpoints
- https://modal.com/docs/guide/sandboxes
- https://modal.com/legal/terms
- https://modal.com/legal/privacy-policy
- https://github.com/modal-labs/modal-client
- https://github.com/modal-labs/modal-client/blob/main/LICENSE
Machine-readable
- JSON
- Markdown
- Source file on GitHub (edit via pull request)