Static Spaces are free for everyone. Creating CPU-backed Gradio/Docker compute Spaces normally requires a paid account tier, even though CPU Basic hardware itself lists a zero hourly rate; limited eligible ZeroGPU exceptions exist. GPUs, model inference and persistent compute are not part of this static tier.

Gemini Developer API Free eligibility, per-model token billing and rate limits vary by current model, geography and request mode. No broad promise of unlimited free API use; free-tier content may be used to improve Google products. Cloud billing is separate for paid-tier access.

Free neuron allowance covers only eligible Workers AI models; requests above it fail until reset or upgrade to Workers Paid, which bills excess neurons. Some frontier models require Workers Paid or prepaid AI Gateway credits even before daily free allocation is exhausted. Standard Workers request/CPU quotas also apply.

Starter is $0/month but usage above monthly allowances is pay-as-you-go: $10/100K observability events, $0.35/CPU hour and $0.10/GB egress. Set credit top-up/auto-top-up cautiously; a persistent always-on server is $100/project. Model/Gateway tokens, storage and other metered services have separate charges. Card requirement and commercial-use terms remain unconfirmed.

Usage limits and short history retention; no credit card required.

Free tier ingests up to 10 logs/minute; API access, alerts, and longer retention require paid plans. Homepage mentions a 7-day trial for premium features; card requirement and commercial-use rights for Hobby need confirmation.

Composio-managed apps have smaller included allowances (20.000 tool calls and 10.000 trigger events/month); third-party provider and Instant Tool charges may apply separately; Hobby pauses at its usage cap. Commercial-use terms not independently confirmed.

Free limits vary by model and organization; exceeding quota returns HTTP 429. Batch/Flex and higher rate limits require paid Developer tier. Check model-specific commercial-use terms.

Cloud data region is US. Additional spans are not included on Free; Pro Cloud is paid. Commercial-use terms for hosted Free are not explicitly confirmed; Ollie coding harness trial is separate from the ongoing Opik Free Cloud allowance.

SaaS Free tier has span and ingestion quotas, short retention, one space and community support. Paid Pro starts at $50/month for higher limits; hosted Free commercial-use terms require confirmation.

Free model availability and provider routing may change; failed requests count toward daily cap. Buying at least $10 credits raises free-model allowance to 1.000 requests/day. Free endpoints may retain prompts; commercial rights and output licenses depend on each underlying model/provider; no SLA.

Each LLM call, tool call, retrieval, evaluation or simulation step counts as an event. Developer has limited retention and evaluation quotas; paid Growth adds capacity. Commercial-use rights for hosted Developer require confirmation.

Requires own compute (typically supported GPU or CPU build), storage and model weights. Infrastructure and model downloads may cost money; each model has separate license and commercial-use restrictions; no free hosted API credits.

You supply and pay for Kubernetes nodes, GPUs/CPUs, storage, network, observability, backups and model downloads; model weights have separate licenses. Knative-based serverless mode requires additional cluster components. No managed inference compute is included.

Operate and secure your own tracking server, database and artifact storage; cloud infrastructure and third-party model/API calls are not included. This is not a managed hosting allowance.

Supply and pay for your own CPU/GPU, network and model storage; model weights may have separate licenses. BentoCloud managed compute is not included; anonymous framework usage telemetry can be disabled.

Requires your own CPU/GPU capacity and Ray deployment; large models, accelerators, storage and external APIs cost extra. Ray Serve is not a fully managed free inference service.

Requires a Kubernetes cluster and substantial infrastructure and operational resources; managed vendor distributions and enterprise support may cost extra. Some experimental subprojects are not production-ready.

Self-hosting, annotation storage and machine-learning backends are your responsibility and may incur costs. Enterprise collaboration, advanced review, governance and managed cloud features are separate paid offerings; this listing is community software only.

The open-source framework and local/self-hosted deployment are distinct from paid ZenML Pro/Cloud offerings. You manage pipeline orchestrators, artifact stores, compute, metadata storage and integrations; external cloud/AI costs apply.

Recommended starting resources are 4 CPU cores, 16 GB RAM and 50 GB disk; Elasticsearch/Infinity and database dependencies require maintenance. LLM/embedding API and infrastructure costs are separate.

Local inference requires your own CPU/GPU, memory, storage and electricity. Individual model weights have their own licenses and commercial-use terms; hosted Ollama cloud models and third-party APIs may have separate limits and charges. MIT license covers the runner, not every model.

Own compute, database and hosting required; third-party model/API calls and vector databases may charge separately. Secure deployments require authentication and patching.

Requires your own GPU/CPU servers and model weights; compute, electricity, cloud GPU and storage may cost money. Model-specific licenses and commercial rights are separate from SGLang's Apache-2.0 code license.

Free limits are small and model-specific; requests beyond the quota are rate-limited. Linking a payment method switches to Developer tier. Preview models are evaluation-only; commercial-use terms require provider confirmation.

Self-hosting requires your own CPU/GPU, RAM, storage and operations; some model weights, runtimes or integrations have separate licenses and costs. Enterprise features and support are separate.

Bring your own hardware, storage and energy; model weights have separate licensing and usage restrictions. Commercial rights for the MIT software do not imply commercial rights for every model.

You supply and pay for hardware, compute, electricity, storage and model weights. Individual model licenses and optional external services may restrict commercial use or add charges; the MIT core license is separate.

enterprise/ code and paid enterprise features require a separate license; upstream LLM API tokens and self-hosted infrastructure are not free.

Models, inference APIs, vector stores and hosting are not included and may incur charges; commercial deepset products are separate.

Modified Apache-2.0 Dify Open Source License: multi-tenant workspace hosting requires commercial authorization, and frontend logos/copyright must remain intact. Minimum documented setup 2 CPU cores and 4 GiB RAM; infrastructure, external models and APIs cost extra. Distinct from Dify Cloud.

Model weights and datasets have their own licenses, access gates and commercial restrictions; check each model separately. Local CPU/GPU, storage, cloud hosting and optional paid model/API endpoints are not included.

Local training/inference needs compatible hardware, GPU memory and model weights; cloud GPUs, hosted inference and third-party model APIs may charge separately. AGPL-3.0 Studio and model-license obligations apply.

Run inference on your own devices or infrastructure; model weights, GPU/accelerator hardware, cloud compute and third-party services may cost extra. Some distribution builds collect usage telemetry; model licenses are separate.

Run and pay for your own CPU/GPU machines, model weights, networking and storage; no hosted inference credits included. Some optimized backends, hardware drivers and NVIDIA AI Enterprise services have separate licenses or costs.

Compute hardware, model licenses, electricity and any optional cloud hosting are separate; some accelerators need supported drivers. OpenVINO telemetry can be opted out; no free hosted model API is included.

You operate the feature server and select online/offline stores; cloud warehouses, Redis/databases, infrastructure and managed services may incur separate charges.

Self-hosting requires your own compute, storage and operations. Evidently Cloud, team collaboration, alerting and enterprise features are separate hosted/paid offerings; optional external LLM judge APIs can incur charges.

DVC CLI is free software, not free hosted object storage or compute. Optional AWS S3, GCS, Azure, remote servers and Git hosting may charge for storage, requests, egress or compute; DVC Studio is a separate offering.

Cloud-scale execution on AWS, Azure, Kubernetes or other providers requires separately provisioned compute, object storage, orchestration and associated charges. This listing covers the free local framework, not managed cloud compute.

You must operate your own server, storage, agents, GPU/CPU resources and dependencies; ClearML's separately hosted or enterprise services may have different pricing and feature limits.

Model inference, hosted endpoints, E2B/Modal/Blaxel sandboxes and compute may charge separately. Code execution requires careful sandboxing and security review; the free library does not include free hosted LLM tokens.

The framework does not include free hosted LLM inference, embedding APIs, vector databases or cloud infrastructure; selected model providers and LlamaCloud hosted services can bill separately. Configure local models and storage for an entirely self-operated setup.

Local CPU/GPU, memory, model downloads and storage are required; external inference APIs or hosted infrastructure may charge. Individual model weights can have different licenses and commercial-use restrictions; verify each model separately.

Pretrained models have independent licenses and possible gated downloads; verify model commercial-use terms separately. Local CPU/GPU resources, hosted inference endpoints, training and cloud infrastructure are not included.

Model checkpoints have separate licenses, usage restrictions and sometimes gated access; check each model before commercial use. GPUs, cloud compute, model downloads and hosted inference APIs are not free with this library.

Bring your own Postgres, Redis, server resources and AI provider API keys; provider inference and infrastructure may incur charges. Hosted AcruxCore is a separate offering. Production deployment requires configured secrets, database and reverse proxy; no free hosted capacity is included.