GroqCloud
Always-free quota
Inputs and outputs are not retained by default or used for training without permission; limited abuse and feature-specific retention exceptions remain.
At least $70.20/moEquivalent free value
Free model APIs.
Prototype freely.
Find hosted LLMs, embeddings, speech, image, and research APIs you can actually use—clearly separated from trials and marketing fluff.
Verified Aug 21, 2026 · 38 ongoing · 76 tracked
Current snapshotYAML → static site
- Ongoing or recurring
- 38
- Community / conditional
- 7
- Trials / promotions
- 31
- Current offers
- 76
- Under review
- 20
- No free offer found
- 37
Every classification links back to current provider documentation, pricing, or a live catalog.
Useful before your first invoice
High-confidence, recurring offers with practical model access. Limits still apply.
Google AI Studio / Gemini Developer API
Always-free quota
Unpaid-service inputs and outputs may be retained and used to improve and train Google models.
Up to $3,640.42/moEquivalent free value
Cloudflare Workers AI
Always-free quota
Workers AI excludes Customer Content from storage, training, and improvement unless the customer deliberately enables a feature or consents.
$3.35/moEquivalent free value
OpenRouter
Free model routes
OpenRouter defaults are protective, but the selected upstream model provider and optional logging or data-sharing controls determine end-to-end handling.
Up to $1,615.75/moEquivalent free value
Free does not mean private.
A provider may retain prompts, permit human review, use free-tier content to improve products or models, or route requests through another company. Those rules can differ between free, paid, and enterprise plans.
How we audit data agreementsFind your free inference route
Search provider names and exact model IDs, then narrow by the kind of free access you need.
a recurring allowance the provider funds indefinitely.specific model routes currently listed at zero price; the set rotates.credits that refresh on a published schedule.no-cost access for research, evaluation, and development — not production use.a one-time allowance for new accounts; it does not recur.community-supplied or end-user-funded capacity with no guaranteed throughput.
GroqCloud
Always-free quotaLlamaQwenGPT OSS12 models
Inputs and outputs are not retained by default or used for training without permission; limited abuse and feature-specific retention exceptions remain.
At least $70.20/mo
Hetzner Experiments Inference API
Experimental accessQwen2 models
Hetzner documents no ordinary prompt/output retention, no content training right, and EU-hosted processing.
Up to $57,354.89/mo
Cloudflare Workers AI
Always-free quotaGemmaLlamaQwen7 models
Workers AI excludes Customer Content from storage, training, and improvement unless the customer deliberately enables a feature or consents.
$3.35/mo
OpenRouter
Free model routesGemmaNemotronGLM22 models
OpenRouter defaults are protective, but the selected upstream model provider and optional logging or data-sharing controls determine end-to-end handling.
Up to $1,615.75/mo
Jina Reader and anonymous utility APIs
Always-free quotaWebpage To MarkdownPdf Extraction2 models
Jina excludes model training, but anonymous Reader retention and non-training content handling remain unclear.
Up to $438.30/mo
AwanLLM
Always-free quotaLlama6 models
AwanLLM says prompts and generations are not logged and discloses no content-based training or improvement use.
At least $79.79/mo
IBM watsonx.ai Runtime
Always-free quotaModel inference4 models
IBM says unsaved API prompts and outputs are not accessed, logged, stored, trained on, or used for improvement without permission.
At least $15.11/mo
Modal
Monthly creditsText generationCustom models
Modal does not store ordinary endpoint payloads or train on Customer Data without written consent.
$30.00/mo
Nous Research / Nous Portal
Free model routesText generation7 models
No usable public policy was found for prompt retention, training, operator access, or deletion.
Up to $21,038.40/mo
Arli AI
Always-free quotaModel inference
Arli AI documents transient-only processing with no prompt/output storage, training, or content-based improvement.
At least $2.33/mo
ModelScope API-Inference
Research / developmentQwen1 model
API-specific retention and training protections are not documented clearly enough for a stronger classification.
At least $652.28/mo
Cohere
Research / developmentCommand10 models
Content can be logged for 30 days and reviewed for abuse, while training is opt-in and aggregate improvement uses remain.
Up to $350.72/mo
Lightning AI Model APIs / litAI
Monthly creditsGeminiGemmaDeepSeek7 models
Model-API retention, training, publisher access, and plan-specific protections are not documented fully.
At least $15.00/mo
Sail Research
Monthly creditsGemmaDeepSeekGPT OSS6 models
Sail limits persistent content storage to 48 hours and excludes content from training and service improvement.
$5.00/mo
Z.AI Model API
Free model routesGLM3 models
The API DPA describes real-time processing without storage and requires explicit agreement before content-based development or improvement.
—
Google AI Studio / Gemini Developer API
Always-free quotaGemini11 models
Unpaid-service inputs and outputs may be retained and used to improve and train Google models.
Up to $3,640.42/mo
Vercel AI Gateway
Monthly creditsText generation1 model
The gateway deletes content after each request, but upstream retention and non-enterprise protections depend on route controls and provider terms.
At least $5.00/mo
Maritaca AI Academic Credits
Research / developmentText generation10 models
No public academic-API agreement defining content retention, training, access, or deletion was found.
At least R$36,817.20/mo
Api.Airforce
Free model routesModel inference
Content can be retained for safety without a fixed TTL, while training and upstream handling are not fully documented.
Up to $456.30/mo
Hugging Face Inference Providers
Monthly creditsText generation
Hugging Face does not store routed bodies, but each selected upstream provider has its own independent data policy.
$0.10/mo
LLM7
Always-free quotaText generation2 models
No usable public policy was found for request-content handling.
Up to $1.83/mo
Requesty LLM Gateway
Free model routesGemmaMistralNemotron12 models
Free self-service traffic is logged by default and can use upstream models that train on content.
Up to $1,436.22/mo
Mistral Studio / La Plateforme
Monthly creditsMistral8 models
Free-mode inputs and outputs are eligible for model improvement and training by default unless the user opts out.
At least $10.00/mo
ElevenLabs API
Always-free quotaText To SpeechSpeech To Text2 models
Non-enterprise API data is retained and eligible for audio-model improvement by default unless the user opts out.
Up to $1.00/mo
NVIDIA Build / NVIDIA-hosted NIM
Research / developmentGemmaLlamaDeepSeek102 models
The free developer terms permit storage and broad product, service, and underlying-model improvement uses.
Up to $4,263.78/mo
OVHcloud AI Endpoints
Free model routesQwen7 models
Endpoint-specific prompt retention and training commitments were not found.
At least $200.00 once
BazaarLink
Free model routesQwen2 models
Prompt TTL, route-specific training, product-improvement scope, and upstream handling remain unresolved.
—
Cartesia
Always-free quotaText To SpeechSpeech To Text2 models
Default terms permit indefinite storage and perpetual use of inputs and outputs for labeling, improvement, promotion, and model training.
Up to $1.00/mo
OrcaRouter
Free model routesQwenDeepSeek4 models
OrcaRouter says it does not retain or train on content, but independently governed upstream providers receive the requests.
—
NSF National Deep Inference Fabric
Research / developmentLlamaGPT OSS4 models
No endpoint-wide public prompt-retention, training, operator-access, or deletion policy was found.
—
OpenTyphoon API
Research / developmentModel inference4 models
No public service agreement defining inference retention, training, access, or deletion was found.
—
Beam
Monthly creditsText generationCustom models
Beam permits storage and expressly permits use of customer data and queries to measure and improve the service.
$30.00/mo
Argonne Leadership Computing Facility Inference Endpoints
Research / developmentGemmaLlamaMistral7 models
Institutional controls apply, but endpoint-wide content retention, training, administrator access, and deletion terms are not public.
—
Kilo AI Gateway
Free model routesNemotronCommand15 models
The terms grant broad perpetual improvement rights and some routed models can require training permission.
—
LLM.API
Free model routesText generation1 model
The gateway defaults to zero retention and no training, but upstream providers are independently governed and optional logging changes retention.
—
DreamPrompting
Always-free quotaGeminiGemmaLlama7 models
Prompts can be stored or publicly shared and user content may be used to improve the service without a gateway-specific no-training promise.
Up to $152.19/mo
OCI Generative AI
One-time trialText generation
Ordinary inference is not stored or used for general model or service improvement; persistence is customer-selected.
$300.00 once
Cerebras Inference
One-time trialGemmaGPT OSS2 models
Cerebras states inference inputs and outputs are not retained or used for content-based training or improvement.
$5.00 once
Scaleway Generative APIs
One-time trialQwenDeepSeekMistral6 models
Scaleway excludes ordinary prompt collection and model training, hosts models on its own EU infrastructure, and limits incident retention.
Up to €5.68 once
AI Horde
Community capacityGemmaLlamaQwen26 models
The coordinator is transient, but volunteer workers can inspect or retain plaintext requests outside a uniform contractual control.
—
Deepgram
One-time trialSpeech To TextText To Speech
Per-request opt-out can exclude training, but participating accounts permit sampled retention and model improvement.
$200.00 once
Fikra API
One-time trialText generation3 models
Fikra states prompts and responses are processed in memory, immediately discarded, excluded from logs, and never used for training.
$0.15 once
Novita AI
One-time trialModel inference
Default inference content is transient and excluded from training and service improvement, subject to narrow legal and support exceptions.
—
Pollinations.ai
Free model routesGemmaGPT OSSGLM9 models
Current free-route retention, training, and multi-path infrastructure handling are not documented precisely.
—
Tencent Hunyuan
One-time trialEmbeddingsImage / vision6 models
Hunyuan free-tier retention, training, improvement, and operator-access commitments were not verified.
Up to CN¥10.30 once
BytePlus ModelArk
One-time trialText generationImage / vision6 models
Campaign terms establish free tokens but do not resolve content TTL, training, improvement, or model-specific routing.
At least $2.40 once
ch.at
Community capacityText generation
The service claims no logs, but the current upstream model and its data handling are not fully documented.
—
Jina Search Foundation API
One-time trialEmbeddingsMultimodal Embeddings
Jina excludes model training on request content, but retention and broader support or improvement handling remain imprecise.
Up to $0.50 once
Speechmatics
One-time trialSpeech To TextText To Speech
Training is opt-in, but batch content is retained for seven days and portal or support features add storage and access.
$100.00 once
Waterfall
Free model routesGemmaNemotronGLM28 models
No usable public inference-data policy was found.
—
AI21 Studio
One-time trialModel inference
Default retention and training treatment are not clear; stronger traceless operation requires explicit configuration.
$10.00 once
Fireworks AI
One-time trialModel inference
Open-model requests are transient and no-training by default, but stored-response and proprietary partner routes can differ.
$1.00 once
Google Cloud Vertex AI
One-time trialModel inference
Google Cloud excludes training without permission, but caching, grounding, and session features introduce bounded retention.
$300.00 once
Puter.js AI
User paysText generation
Puter and the selected upstream both process requests, while AI-specific TTL and training protections are unresolved.
—
WaveSpeedAI
One-time trialLanguage ModelsImage
A no-training statement exists, but prompt/output TTL, deletion, and third-party model handling remain unresolved.
$1.00 once
Amazon Bedrock
One-time trialText generation
Bedrock is zero-retention and no-training by default, but named model routes have current abuse-review retention exceptions.
At least $100.00 once
Replicate
One-time trialModel inference
API data is deleted after one hour by default, but community models, resultant data, and web history introduce additional handling.
—
FreeInference.org
Always-free quotaGemmaQwenDeepSeek7 models
Prompts and responses may be logged and reused for service research, improvement, and published derived datasets.
—
Inception Platform
One-time trialText generation2 models
No usable public inference-data policy was found.
Up to $75.00 once
Microsoft Foundry Models
One-time trialModel inference
Base inference is stateless and no-training, but abuse review and stateful features can retain content for human access.
$200.00 once
Nebius Token Factory
One-time trialModel inference
Training is excluded, but prompts and outputs may be stored by default unless zero-data-retention mode is enabled.
$1.00 once
Poolside direct API
Limited promotionModel inference4 models
No usable public inference-data policy was found for the promotion.
—
Alibaba Cloud Model Studio / Qwen API
One-time trialModel inference
Alibaba excludes training without consent, but request data is stored regionally and third-party model terms can differ.
—
Hyperbolic
One-time trialModel inference
The AI FAQ promises transient no-training inference, but feedback storage, node operators, and a broader content license remain.
$1.00 once
Mancer AI
Free model routesText generation1 model
Default terms permit retention, third-party disclosure, model training, and service improvement unless the user opts out.
—
Public AI Inference Utility
One-time trialGemmaQwen11 models
No usable public inference-data agreement was found.
—
MARA Inference Cloud
One-time trialGemmaDeepSeekGPT OSS5 models
No usable public inference-data policy was found.
$5.00 once
Sarvam AI
One-time trialGemmaGLM3 models
Free developer inputs and outputs are collected and may support service improvement without a clear default no-training commitment.
₹100.00 once
Logfare
Always-free quotaGemmaQwenDeepSeek20 models
Every request body is logged and scrubbed content may be used in internal evaluation datasets; premium access can require training opt-in.
—
FastRouter
Free model routesGemmaGPT OSSNemotron5 models
The terms grant broad perpetual rights to store and use inputs and outputs for debugging and service improvement.
—
Voyage AI
One-time trialEmbeddingsMultimodal Embeddings4 models
Standard terms permit perpetual storage, model training, and improvement unless an eligible organization opts out.
At least $114.00 once
AssemblyAI
One-time trialSpeech To TextSpeech Understanding
Standard terms allow indefinite retention in some modes and permit training and broad improvement unless excluded or opted out.
$50.00 once
Upstage
One-time trialModel inference
Current free-service terms expressly allow storage, service improvement, AI research, and model training with inputs and outputs.
At least $10.00 once
Clarifai
One-time trialModel inference
Inputs and predictions are stored by default and the general terms retain broad service-development rights.
$5.00 once
OpenCode Zen
Limited promotionDeepSeekNemotron9 models
Named free routes may collect prompts and completions for model improvement or training, with route-specific retention.
—
SEA-LION API
One-time trialGemmaLlamaQwen5 models
The terms authorize use of inputs and outputs to develop and improve the service without a clear no-training commitment.
—
“Start free” is not always free inference.
We keep disputed offers on a watchlist and document providers whose free tier is retired, UI-only, BYOK, or simply paid.