llms.txt
Compare with other sites' llms.txt files →# FlexAI
> FlexAI is an AI cloud platform that makes it easier to build, run and
> scale AI, from models to agents, with freedom to choose your models,
> runtimes, hardware and where your workloads run. One account spans model
> inference through Token Factory, agent execution through AgentOS (in
> trial), VMs, fine-tuning and production inference through Dedicated
> GPUs,
> and private infrastructure through AI Factory.
Customer data: FlexAI applies zero data retention to all customer content in
every workspace, Global and EU (prompts, inputs and outputs are not stored
after a request is processed), and never uses customer data to train,
fine-tune or evaluate any model. See https://flex.ai/privacy-policy.
This file follows the llms.txt convention (https://llmstxt.org/) so that
answer-engine and LLM crawlers can discover the canonical pages on flex.ai
without parsing the marketing chrome.
A companion machine-readable catalog lives at https://flex.ai/llms-full.txt:
every served model with its canonical URL, availability, modality and current
public rates. It is an experimental convenience for crawlers, not a ranking or
a recommendation.
## Products
- [Token Factory](https://flex.ai/token-factory): OpenAI-compatible inference API. One key across text, vision, image, video, and audio models, priced by usage (per token for text and vision; per image, per minute of generated video, or per minute or million characters for audio).
- [Dedicated GPUs](https://flex.ai/dedicated-gpus): VMs, fine-tuning and production inference on NVIDIA and AMD GPUs that serve only your traffic, on the same OpenAI-compatible account.
- [Models](https://flex.ai/models): The live model catalog: serverless (Token Factory) and dedicated-endpoint open models with per-model pricing and specs.
- [Inference](https://flex.ai/inference): Serverless, batch, and dedicated inference endpoints with auto-scaling; dedicated GPUs are priced per GPU-hour, metered per minute, and autoscale.
- [AgentOS](https://flex.ai/agent-os): Run your agents on FlexAI with runtime, model, and hardware independence. Available in trial for selected teams.
- [Fine-tuning](https://flex.ai/fine-tuning): Managed fine-tuning with LoRA/QLoRA on open-source models.
- [Training](https://flex.ai/training): Distributed training jobs with checkpoint resume and multi-node DDP.
- [Platform overview](https://flex.ai/platform): How the managed AI services and infrastructure building blocks fit together.
- [AI Factory](https://flex.ai/ai-factory): FlexAI's AI services deployed on your own hardware: VPC, on-prem, or air-gapped. Design partners: talk to us.
## Use cases
- [Use cases](https://flex.ai/use-cases): Common agents built as multi-model pipelines behind one key.
- [Coding agents](https://flex.ai/use-cases/coding-agents): Generate, review, and repair code (Qwen3 Coder, GPT-OSS).
- [Research agents](https://flex.ai/use-cases/research-agents): Retrieve, reason, and synthesize (BGE-M3, DeepSeek).
- [Support agents](https://flex.ai/use-cases/support-agents): Transcribe, retrieve, respond, and speak (Whisper, BGE-M3, GPT-OSS, Kokoro).
- [Workflow automation](https://flex.ai/use-cases/workflow-automation): Plan and execute multi-step tool calls (GPT-OSS, Mistral).
- [Multimodal generation](https://flex.ai/use-cases/multimodal-generation): Generate images and describe them (FLUX.1, Gemma).
- [Private copilots](https://flex.ai/use-cases/private-copilots): Grounded assistants over your own data (BGE-M3, Qwen, Llama).
- [Account briefs](https://flex.ai/use-cases/account-briefs): How FlexAI's sales team preps every account meeting with Maestro on AgentOS (in trial): HubSpot, Slack, Google Calendar and Drive, Notion and web search into a sourced brief, with scoped access to each rep's own data.
## Model families
- [Qwen on FlexAI](https://flex.ai/models/qwen): Every served Qwen variant: serverless behind one key, dedicated on your own endpoint.
- [DeepSeek on FlexAI](https://flex.ai/models/deepseek): Every served DeepSeek variant: serverless behind one key, dedicated on your own endpoint.
- [Mistral on FlexAI](https://flex.ai/models/mistral): Every served Mistral variant: serverless behind one key, dedicated on your own endpoint.
- [Nemotron on FlexAI](https://flex.ai/models/nemotron): Every served Nemotron variant: serverless behind one key, dedicated on your own endpoint.
- [Gemma on FlexAI](https://flex.ai/models/gemma): Every served Gemma variant: serverless behind one key, dedicated on your own endpoint.
- [GLM on FlexAI](https://flex.ai/models/glm): Every served GLM variant: serverless behind one key, dedicated on your own endpoint.
- [Llama on FlexAI](https://flex.ai/models/llama): Every served Llama variant: serverless behind one key, dedicated on your own endpoint.
- [GPT-OSS on FlexAI](https://flex.ai/models/gpt-oss): Every served GPT-OSS variant: serverless behind one key, dedicated on your own endpoint.
## Methodology
- [How FlexAI pricing is set](https://flex.ai/pricing#how-pricing-works): the rule behind every serverless rate: each model is priced at the credible market rate and repriced automatically when the source moves. Published so a rate can be checked rather than taken on trust.
- [Serverless vs dedicated break-even](https://flex.ai/tools/serverless-vs-dedicated): where per-token billing stops being cheaper than a dedicated endpoint, per model.
- [GPU savings calculator](https://flex.ai/tools/savings-calculator): compare GPU-hour costs against eight providers. Scenarios built from published provider rates are shareable URLs, so a result can be reproduced; scenarios using your own negotiated pricing stay local and are deliberately not encoded in the link.
## Capabilities
- [Embeddings](https://flex.ai/models/embeddings): Open embedding models on the OpenAI-compatible /v1/embeddings endpoint, billed per token.
- [Speech-to-text](https://flex.ai/models/speech-to-text): Open transcription models on /v1/audio/transcriptions, billed per minute.
- [Text-to-speech](https://flex.ai/models/text-to-speech): Open TTS models on /v1/audio/speech, billed per character.
- [Image generation](https://flex.ai/models/image-generation): Open image models on /v1/images/generations, billed per image.
- [Video generation](https://flex.ai/models/video-generation): Open text-to-video (Wan2.2) on /v1/videos/generations (async), available on demand on dedicated GPUs.
## Pricing & comparison
- [Pricing](https://flex.ai/pricing): Serverless per-token rates (Token Factory), dedicated per-GPU-hour rates, and custom private-cloud tiers.
- [Token Factory savings calculator](https://flex.ai/tools/open-source-llm-cost-calculator): Compare paying frontier-model APIs (OpenAI, Anthropic, Google) vs. equivalent-quality open models on FlexAI Token Factory.
- [Savings calculator](https://flex.ai/tools/savings-calculator): Estimate GPU-compute spend reduction from migrating workloads to FlexAI.
- [Serverless vs. dedicated](https://flex.ai/tools/serverless-vs-dedicated): Find the token volume where a dedicated endpoint beats per-token pricing.
- [FlexAI vs AWS Bedrock](https://flex.ai/vs/aws-bedrock)
- [FlexAI vs OpenAI](https://flex.ai/vs/openai)
- [FlexAI vs Fireworks](https://flex.ai/vs/fireworks)
- [FlexAI vs CoreWeave](https://flex.ai/vs/coreweave)
- [FlexAI vs Together](https://flex.ai/vs/together)
## Case studies
- [LegML](https://flex.ai/case-studies/legml): A 32B French legal LLM fine-tuned on FlexAI that outperformed a frontier model at lower cost.
- [DragonLLM](https://flex.ai/case-studies/dragonllm): Private inference for finance, hosted in France. Autoscaling, scale-to-zero.
- [Pixelcut](https://flex.ai/case-studies/pixelcut): Image-generation fine-tuning with pay-per-use economics.
## Blog & technical content
- [Engineering](https://flex.ai/engineering): The engineering pillar: benchmarks, heterogeneous compute, training and RL, and the agent harness.
- [Blog](https://flex.ai/blog): Engineering and product writing on AI infrastructure.
- [Blueprints](https://flex.ai/blueprints): Reference recipes for common workloads (RAG, multi-agent, text-to-video, RL fine-tuning).
- [Documentation](https://docs.flex.ai): Full developer documentation, API reference, and runtime guides.
## Company
- [Why Us](https://flex.ai/why-us): From making compute fluid to managed inference. Why FlexAI leads with Token Factory.
- [Growing AI teams](https://flex.ai/companies): FlexAI for growing AI teams, scale production AI on one account.
- [Enterprise](https://flex.ai/enterprise): Private AI cloud: VPC, on-prem, and air-gapped deployment paths.
- [System Integrators](https://flex.ai/system-integrators): Resell and white-label FlexAI managed AI infrastructure to your customers.
- [Partnerships](https://flex.ai/partnerships): Work with FlexAI as a compute, model, or technology partner.
- [Cloud Providers](https://flex.ai/cloud-providers): For GPU clouds and datacenter operators: license AI Factory and offer Token Factory APIs and Dedicated GPUs to your customers.
- [Startup Program](https://flex.ai/startups): Credits, discounts, and onboarding for early-stage AI companies.
- [Careers](https://flex.ai/careers): Open roles.
- [Contact](https://flex.ai/contact): Sales, partnerships, and general inquiries.
## Legal
- [Privacy policy](https://flex.ai/privacy-policy)
- [Terms of service](https://flex.ai/terms-of-service)
- [Acceptable use policy](https://flex.ai/acceptable-use-policy)