distil labs
distil labs provides a platform that trains, deploys, and hosts custom task-specific small language models from production traces, enabling developer teams and enterprises to replace frontier LLM API calls with lower-cost, task-tuned SLMs served through an OpenAI-compatible endpoint.
- Company typePrivate
- Founded-
- HeadquartersBerlin, Germany
- Headcount1–10
- GTM typeB2B
- OfferingSoftware
What distil labs does
distil labs (Distil Labs GmbH, Berlin) provides a platform for training, deploying, and hosting custom task-specific small language models (SLMs) from production traces. The core offering automates knowledge distillation through a six-stage pipeline — synthetic data generation from traces or seed examples, validation and deduplication, distribution matching, supervised fine-tuning with reinforcement learning, 16-to-4-bit quantization, and optimized deployment compilation — and exposes the resulting models behind an OpenAI-compatible API endpoint that customers integrate by changing their base URL. Proprietary techniques include 'vibe-tuning,' a prompt-based fine-tuning method that converts natural language task descriptions into deployable SLMs in hours, and a synthetic-data-from-traces methodology that internal benchmarks show scoring up to 26 percentage points higher than direct trace training. The January 2025 launch of vibe-tuning is the most recent product expansion. The company monetizes through usage-based GPU compute fees, pre-paid training credits (12-month expiry), and per-token fallback inference charges, with a free first-model training as a PLG entry point. Customer base spans regulated on-prem deployments (Rocketgraph on IBM Power), high-volume EdTech content classification (Knowunity processing hundreds of millions of monthly requests), cybersecurity agents (Octodet's KINDI), and developer-led SaaS (Uptime Industries, MacPaw), and the company reports that 30M+ end users interact with distil labs-trained models.
distil labs pursues a hybrid go-to-market: self-serve PLG via app.distillabs.ai for developers and SMBs, direct enterprise sales via calendly for regulated and on-premises deployments, and a recently launched Inference Implementation Partner program that recruits GPU infrastructure providers and cloud resellers to resell custom-trained models to their own customers. Strategic partners include Project A Ventures, NAP VC, and helloworld VC as financial backers, and AWS, Google, NVIDIA, and IBM Power as technology or co-development partners, with IBM Power representing a flagship co-marketing relationship for private AI on enterprise hardware. The company has 8 employees, is incorporated as a private limited company in Germany, and does not yet hold formal SOC 2 or ISO 27001 certification though it states alignment to SOC 2 guidelines.
distil labs firmographics
Firmographics- Name
- distil labs
- Legal name
- Distil Labs GmbH
- Website
- https://distillabs.ai
- Company type
- Private
- Operating status
- Operating
- Headcount range
- 1–10 employees
- Short description
- distil labs provides a platform that trains, deploys, and hosts custom task-specific small language models from production traces, enabling developer teams and enterprises to replace frontier LLM API calls with lower-cost, task-tuned SLMs served through an OpenAI-compatible endpoint.
- Ownership category
- akta.pro rank
distil labs industry classification
Industry- Product category
- Small Language Model Platform
- NAICS
- Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (5182), Computer Systems Design Services (541512)
- SIC
- Services-Prepackaged Software (7372)
- akta.pro primary industry
- Model Hosting, Serving & Inference Platforms (HDAAACAB)
- akta.pro secondary industry
- Managed Application Platforms (App PaaS & Runtime Services) (HDABAAAD)
Keywords
Where distil labs is headquartered
LocationHeadquarters
- HQ city
- Berlin
- HQ country
- Germany
- HQ region
- Europe
Offices1 record
Markets served
distil labs business model
Business model- GTM type
- B2B
- Offering type
- Software
- Cost components
- Personnel, Technology or R&D, Infrastructure, Marketing or Sales, Operations
Revenue model
- Usage-based GPU Compute: Customers pay fees calculated based on GPU hours consumed through the platform. Models run on dedicated GPUs with consumption-based pricing that scales with usage. Pricing depends on utilization - higher throughput means lower per-request cost.
- Training Credits: Fees for training credits used for fine-tuning and model updates. Pre-paid training credit packages available with 12-month expiration. No minimum spend commitments unless separately agreed.
- Fallback Inference Fees: Per-token fees charged when fallback inference is triggered due to capacity exceeding dedicated SLM resources.
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Freemium | One-time | Free trial - first model training |
| Usage-based | Pay-as-you-go | Usage-based GPU compute |
| Subscription | Annual | Training Credits |
| Usage-based | Pay-as-you-go | Fine-tuned SLM inference cost |
Go-to-market motion3 records
Distribution channels3 records
Marketing channels6 records
distil labs product offering
Product offeringCore offering
distil labs provides a cloud platform that trains, deploys, and hosts custom task-specific small language models (SLMs) from production traces or seed examples. Customers route a fraction of their LLM traffic through the platform to capture traces, which are transformed via a six-stage knowledge distillation pipeline (synthetic data generation, validation, distribution matching, SFT + RL, quantization, optimized deployment) into production-ready SLMs served behind an OpenAI-compatible API endpoint. The output is a smaller, cheaper model that matches frontier LLM accuracy at roughly 20% of the cost-per-request, supporting both cloud and on-premises deployment.
Product overview
distil labs is a unified platform for training and deploying task-specific small language models (SLMs) using knowledge distillation. The core offering combines the distil labs Platform with a six-stage SLM Training Pipeline that automates synthetic data generation, validation, fine-tuning, and deployment. Customers interact primarily through OpenAI-Compatible Endpoints, enabling seamless API replacement. The platform offers Vibe-Tuning for prompt-based fine-tuning and Production Trace Collection for capturing real training data. Models are 50-400x smaller than frontier LLMs while matching accuracy at 80% lower cost.
Differentiator
Problem solved
Functional benefit
Products and services
- distil labs Platform A cloud-based platform for training task-specific small language models (SLMs) using knowledge distillation. Customers provide a task description and examples or route production traffic through the platform; it handles synthetic data generation, validation, fine-tuning, evaluation, and deployment to deliver production-ready SLMs that match frontier LLM accuracy at roughly 20% of the cost. Targets engineering and AI platform teams.
Quantifiable outcome
- 80% lower cost per request vs frontier LLMs
- +4 more outcomes
Companies that use distil labs
Customer profileNamed customers6 records
Segments5 records
Ideal customer profiles5 records
distil labs technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- Yes
- API docs
- API detail
Core technology
AI maturity
App detail
AI capability5 records
Feature5 records
distil labs partnerships and signals
Strategic signalPartnerships
Five partnerships are on record, tiered infrastructure, flagship and growth.
- Amazon Web ServicesinfrastructureCloud infrastructure provider for hosting the distil labs platform. Platform operated on established third-party cloud infrastructure providers selected by distil labs.
- GoogleinfrastructureCloud infrastructure provider potentially used for platform operations. Listed in 'Backed by' section alongside other infrastructure partners.
- NVIDIAinfrastructureGPU provider for model training and inference. H100 GPUs used for benchmark testing. Listed in 'Backed by' section indicating partnership.
- IBM PowerflagshipCo-development of private AI solutions on IBM Power infrastructure. Unnikrishnan Rajagopal, Director at IBM Power, stated: 'The collaboration with distil labs shows what's possible when we fine-tune small language models and deploy them on IBM Power. It's about delivering private, accurate AI that's ready for enterprise scale.' Rocketgraph is a joint customer case study.
- GPU Infrastructure PartnersgrowthInference Implementation Partner program targeting GPU providers, cloud resellers, and infrastructure companies. Partners unlock stuck deals, migrate from serverless to dedicated endpoints, and monetize small-GPU capacity by offering custom model training to their customers.
Scale indicators7 records
Recent moves6 records
Expansion highlights6 records
distil labs competitors and assessment
Company assessmentBroad incumbents
- Hugging Face: Broad AI platform offering model hosting, inference endpoints, fine-tuning (AutoTrain, Spaces), and a large model marketplace. Overlaps with distil labs on custom-model deployment but as part of a much wider portfolio.
- Cohere: Enterprise-focused foundation model provider offering fine-tuning, custom models, and private deployment. Competes in regulated-industry verticals distil labs targets via Rocketgraph/IBM Power partnerships.
Direct peers
- Modal Labs: Developer-focused cloud platform for running and serving AI models and workloads on GPU infrastructure. Targets similar developer/ML practitioner audience with self-serve deployment and inference offerings.
- Anyscale: AI compute platform built on Ray for training, serving, and fine-tuning LLMs and other models at scale. Comparable in offering custom-model hosting with cost-optimized inference infrastructure.
- Fireworks AI: Model inference and fine-tuning platform optimized for low-latency, cost-efficient deployment of custom and open-source models. Competes head-to-head on cost-per-token economics and OpenAI-compatible APIs.
- Lamini: Enterprise platform specialized in fine-tuning and serving custom LLMs for production workloads. Direct overlap in custom-model distillation, training pipelines, and dedicated inference infrastructure.
- Predibase: Developer platform for fine-tuning and serving open-source LLMs with built-in evaluation and deployment tooling. Targets similar cost-conscious engineering teams seeking to replace frontier API spend.
- Replicate: Cloud platform for running open-source and custom ML models via API. Offers fine-tuning and deployment flows targeting developers building production AI features, similar to distil labs' self-serve motion.
- Together AI: Cloud platform for training, fine-tuning, and serving open and custom AI models at scale. Direct competitor offering dedicated GPU inference, fine-tuning APIs, and OpenAI-compatible endpoints — highly comparable business model and customer overlap with distil labs.
Emerging players
- Prem AI: Privacy-focused platform for fine-tuning and hosting custom LLMs, with emphasis on data control and self-hosting. Comparable niche overlap with distil labs' regulated-industry and on-premises positioning.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat5 records
Key risks6 records
Key highlights7 records
Customer concentration
distil labs social profiles
Digital presencedistil labs compliance and trust
Trust signalCompliance1 record
distil labs financial estimates
Financial estimateRevenue estimate
Valuation estimate
distil labs leadership team
Management profileNumber of profiles
Profiles8 records
distil labs funding detail
Funding detailFunding overview
Funding rounds
Investors
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
distil labs M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about distil labs
What does distil labs do?
distil labs provides a cloud platform that trains, deploys, and hosts custom task-specific small language models (SLMs) from production traces or seed examples. Customers route a fraction of their LLM traffic through the platform to capture traces, which are transformed via a six-stage knowledge distillation pipeline (synthetic data generation, validation, distribution matching, SFT + RL, quantization, optimized deployment) into production-ready SLMs served behind an OpenAI-compatible API endpoint. The output is a smaller, cheaper model that matches frontier LLM accuracy at roughly 20% of the cost-per-request, supporting both cloud and on-premises deployment.
Is distil labs a public or private company?
distil labs is a private company. It is classified as venture growth investor backed and is currently operating.
When was distil labs founded?
distil labs was founded in -1. It employs 1 to 10 people.
Where is distil labs based?
distil labs is headquartered in Berlin, Germany, in the Europe region.
How does distil labs make money?
Three revenue lines are on record. Usage-based GPU Compute is the primary driver. The others are training Credits and fallback Inference Fees.
Who are distil labs's main competitors?
Broad incumbents on record are Hugging Face and Cohere. Direct peers are Modal Labs, Anyscale, Fireworks AI, Lamini, Predibase, Replicate and Together AI. Prem AI is listed as an emerging player.
Does distil labs have an API?
Yes. distil labs provides an OpenAI-compatible API endpoint. The platform's trained models are accessible via the Chat Completions API format, allowing integration by simply changing the base URL in OpenAI client configurations. The endpoint supports model deployment and inference serving, with rate limits and access controlled through API keys. Developer documentation is at www.distillabs.ai/docs.
What industry is distil labs in?
distil labs's product category is Small Language Model Platform. Its primary akta.pro industry code is HDAAACAB, Model Hosting, Serving & Inference Platforms, with a secondary code of HDABAAAD, Managed Application Platforms (App PaaS & Runtime Services). Its NAICS code is 5182 and its SIC code is 7372.