What Doubleword does

Doubleword is a London-based AI infrastructure company (legally TYTN LTD) that provides an OpenAI-compatible API for high-throughput asynchronous and batch LLM inference, targeting AI engineering teams whose unit economics are broken by real-time API pricing on background workloads. The platform is built on a proprietary five-layer inference stack consisting of a Rust-based model gateway (Control Layer), intelligent GPU scheduling and orchestration, an optimized runtime engine built on TensorRT-LLM, SGLang, and vLLM with custom techniques including speculative KV coding, Cloudburst cold-start acceleration, and ZeroDP JIT weight offloading, a curated portfolio of open-weight models (DeepSeek, Qwen, Kimi, GLM, Gemma, Nemotron families across text, vision, OCR, and embeddings), and flexible multi-cloud GPU infrastructure that uses spot instances to minimize cost. Customers access three pricing tiers (Realtime, Async at 25-50% off, Batch/Overnight at 50-80% off), with documented cost outcomes of 4-9x below frontier closed APIs on comparable-intelligence workloads.

The company operates a product-led, self-serve go-to-market: developers sign up at app.doubleword.ai, receive an instant API key with no credit card required, and migrate by changing the base URL on existing OpenAI client code. Supporting tooling includes open-source libraries (Autobatcher for transparent request batching, QLM query language), production-ready workbooks on GitHub covering async agents, synthetic data, evals, and ETL pipelines, and named integrations with developer platforms (Arize AI, LangSmith, OpenClaw). Reference customers include OpenMed (medical VQA at 94% below Anthropic), Dataiku's 575 Lab (synthetic PII generation at $50 total cost), and a community of individual developers publishing use cases on LinkedIn, Medium, and Hugging Face.

Doubleword raised a $12 million seed round led by Dawn Capital in May 2025 (with prior funding from Octopus Ventures and Intel Ignite), and in April 2026 was selected as one of six startups to receive up to one million GPU hours on the UK's AI Research Resource supercomputer network under the £500M Sovereign AI initiative. The company is privately held, headquartered in London, and structured around an inference-systems engineering team of 11-50 employees.

Name

Doubleword

Legal name

TYTN LTD

Website

https://doubleword.ai

Company type

Private

Founded year

2021

Operating status

Operating

Headcount range

11–50 employees

Short description

Doubleword is a London-based AI infrastructure company providing an OpenAI-compatible API for high-throughput asynchronous and batch LLM inference, delivering open-weight model serving at up to 80-90% lower cost than real-time frontier APIs for AI engineering teams running background agents, data pipelines, evaluations, and synthetic data generation.

Ownership category

akta.pro rank

Product category

Cloud AI Inference

NAICS

Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (518), Custom Computer Programming Services (541511), Computer Systems Design and Related Services (54151)

SIC

Services-Computer Programming, Data Processing, Etc. (7370), Services-Computer Integrated Systems Design (7373), Services-Computer Programming Services (7371)

akta.pro primary industry

Model Hosting, Serving & Inference Platforms (HDAAACAB)

akta.pro secondary industries

Model Deployment, Serving & Inference Platforms (HDAAABAF), Model Serving, Inference & Deployment Platforms (APIs, Edge/On-Prem) (HDAEANAC), Model Compression & Optimization (Quantization, Distillation, Pruning) (HDAAACAL), Generative AI & LLM Solutions Services (RAG, Agents, Copilots) (BPAEAHAG)

Keywords

• LLM inference API
• Batch AI inference
• Async language models
• AI inference infrastructure
• Open-weight model serving

Headquarters

HQ city

London

HQ country

United Kingdom

HQ region

Europe

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Infrastructure, Technology or R&D, Personnel, Marketing or Sales, Operations

Revenue model

1. LLM Inference API (Usage-Based): Pay-per-token pricing for API access to hosted open-weight LLM models. Three tiers: Realtime (full price, latency-optimized), Async (25-50% discount, high-throughput), and Batch/Overnight (50-80% discount, 24H SLA). No minimum spend, no credit card required. Revenue scales with token volume consumed by customers.

Pricing tiers

Go-to-market motion2 records

Distribution channels3 records

Marketing channels8 records

Core offering

Doubleword operates a high-throughput AI inference API platform delivering hosted access to open-weight large language models (LLMs) at up to 80–90% lower cost than real-time frontier APIs. The platform offers three pricing tiers (Realtime, Async, Batch/Overnight) backed by a proprietary five-layer inference stack — a Rust-based model gateway, intelligent orchestration, an optimized runtime engine (TensorRT-LLM/SGLang/vLLM), curated quantized open-weight models, and flexible multi-cloud GPU infrastructure.

Product overview

Doubleword is a London-based AI inference provider offering a unified platform for high-throughput async and batch inference at up to 90% lower cost than real-time APIs. The core product is the Doubleword API (OpenAI-compatible at https://api.doubleword.ai/v1), which delivers three pricing tiers: Realtime, Async, and Batch/Overnight. This is powered by the Doubleword Inference Stack—a proprietary five-layer infrastructure (gateway, orchestration, runtime engine, models, hardware). The platform provides access to an extensive portfolio of open-weight models including text models (DeepSeek-V4-Pro, Qwen3.5-397B, GLM-5.2, Kimi-K2.6), vision-language models (Qwen3-VL), embedding models (Qwen3-Embedding-8B), and OCR models (DeepSeek-OCR-2, olmOCR-2-7B, LightOnOCR-2-1B). Supporting resources include Workbooks (production-ready templates), the Autobatcher library (drop-in batching), the open-source Control Layer gateway, a Savings Calculator, and an AI Glossary. Integrations with Arize AI, LangSmith, and OpenClaw extend evaluation and agent workflow capabilities.

Differentiator

Problem solved

Functional benefit

Products and services

• Doubleword Inference API Public OpenAI-compatible REST API for high-throughput async and batch inference of open-weight LLMs. Supports Chat Completions, Responses, Embeddings, and Batch endpoints with tool calling, structured JSON generation, and webhook delivery. Three pricing tiers (Realtime, Async 25-50% off, Batch 50-80% off) targeting developers and engineering teams running background agents, evaluations, ETL pipelines, and synthetic data generation.
• Open-Weight Models Portfolio Catalog of open-weight LLMs hosted and served through the Doubleword API, including text generation models (DeepSeek-V4-Pro/Flash, Qwen3.5 family, GLM-5.2, Kimi-K2.6, Gemma-4-31B, Nemotron-3-Ultra-550B, GPT-OSS-20B), vision-language models (Qwen3-VL-235B/30B), embedding models (Qwen3-Embedding-8B), and OCR/document models (DeepSeek-OCR-2, olmOCR-2-7B, LightOnOCR-2-1B). Models are delivered with INT8/FP8/INT4 quantization for cost efficiency.
• Autobatcher Open-source Python library acting as a drop-in AsyncOpenAI replacement that transparently batches requests over a configurable window to capture batch-tier pricing without code rewrites or .jsonl file management. Enables existing async OpenAI code to switch to Doubleword by changing the base URL.
• Control Layer Open-source Rust-based model gateway handling multi-model routing, access controls, logging, and monitoring at scale. Claims 450× less overhead than LiteLLM. Designed for teams self-hosting LLM routing infrastructure.
• QLM (Query Language for Models) Open-source query language tool providing a structured interface for interacting with models hosted on Doubleword's API. Available on GitHub.

Quantifiable outcome

• Up to 90% cost reduction vs Anthropic Claude Opus 4.6 real-time API for comparable intelligence (Qwen3.5-4B Batch: $100 vs Anthropic ~$30K per 1B in+out)
• +8 more outcomes

Named customers6 records

Segments4 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration3 records

AI capability9 records

Feature11 records

Partnerships

Six partnerships are on record, tiered core and secondary.

• Arize AIcoreTechnology or IntegrationIntegration between Doubleword's batch inference API and Arize AX observability platform. Doubleword configured as a Custom Model Endpoint in Arize AX, enabling continuous LLM-as-judge evaluations with OpenTelemetry trace visualization. Frontier judge models (DeepSeek V4 Pro) run at batch cost; Arize handles trace topology and scoring dashboards.
• LangSmithsecondaryTechnology or IntegrationLangSmith integration allowing users to run LLM-as-judge evaluations via Doubleword's batch API. Demonstrated $8.82 for 10K evals — 97% cheaper than GPT-5.5 real-time with zero 429 errors.
• OpenClawsecondaryTechnology or IntegrationOpenClaw computer-use agent platform has an official Doubleword skill enabling background inference for always-on agents. Agents route background tasks (research, email triage, doc processing, cron jobs) to Doubleword async tier at up to 10× lower cost than real-time endpoints. Skill published at github.com/doublewordai/batch-skill.
• OpenMedsecondaryStrategic or Co-development PartnerOpenMed used Doubleword's batch inference API for SynthVision project — 119,137 medical images annotated using Qwen 397B and Kimi K2.5 frontier VLMs at $452.58 total. Resulting 110K medical VQA records open-sourced on HuggingFace. Doubleword serves as the inference infrastructure partner for the research collaboration.
• DataikusecondaryStrategic or Co-development PartnerDataiku's 575 Lab used Doubleword to generate synthetic PII training data for Kiji Privacy Proxy — a privacy protection model for enterprise AI workflows. $50 cost for 26 PII entity types vs ~$1,000 with closed-source providers. Kiji Privacy Proxy and training dataset published under Apache 2.0 on HuggingFace.
• Hugging FacesecondaryStrategic or Co-development PartnerHugging Face hosts the SynthVision dataset (OpenMed × HuggingFace collaboration) — 110K synthetic medical VQA records produced using Doubleword's batch inference. The dataset and resulting model adapters are open-sourced on HuggingFace's platform, with Doubleword credited as the inference infrastructure.

Scale indicators3 records

Recent moves6 records

Expansion highlights6 records

Broad incumbents

• AWS Bedrock: AWS Bedrock is a managed service offering foundation models including open-weight options, bundled with broader AWS infrastructure and enterprise SLAs. It competes with Doubleword on open-weight inference at scale but as part of a much larger cloud portfolio.
• Google Vertex AI: Vertex AI is Google's enterprise AI platform offering model hosting, fine-tuning, and inference including open-weight models. It competes broadly with Doubleword in serving enterprise LLM workloads, particularly for organizations already committed to Google Cloud.
• Azure AI Foundry: Azure AI Foundry (formerly Azure AI Studio) hosts open-weight and proprietary models with enterprise-grade governance and procurement integration. It is a broad incumbent competing with Doubleword for enterprise inference workloads.

Direct peers

• DeepInfra: DeepInfra is a low-cost serverless inference provider for open-weight models, explicitly positioning on price-per-token economics. It is a direct functional competitor to Doubleword's async and batch tiers.
• Together AI: Together AI is a direct competitor offering a cloud platform for open-weight LLM inference and fine-tuning at low per-token pricing. Both target developers seeking cheap, high-throughput access to open-weight models and emphasize cost-per-token as the primary differentiator.
• Replicate: Replicate hosts open-weight AI models via a pay-per-prediction API with strong developer ergonomics. It is comparable to Doubleword in target customer (developers running open models) and pricing model (usage-based), though with broader model coverage beyond LLMs.
• Anyscale: Anyscale (built on Ray) offers managed compute and inference for AI workloads, including open-weight LLM serving. It overlaps with Doubleword's focus on efficient inference at scale and serves similar ML engineering audiences.
• Fireworks AI: Fireworks AI provides a serverless inference platform for open-weight LLMs and custom models, with a strong focus on throughput and low latency. It competes head-to-head with Doubleword on cost-per-token and supports async/batch workloads.

Emerging players

• Modal Labs: Modal provides serverless compute for AI and data workloads, including LLM inference. It overlaps with Doubleword in serving developers running GPU-accelerated jobs but with a more general-purpose compute framing.
• Lepton AI: Lepton AI offers a cloud platform for running and deploying open-weight AI models, with emphasis on developer experience and cost efficiency. It targets similar ML engineering buyers as Doubleword.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights7 records

Customer concentration

• GitHubgithub.com/doublewordai
• LinkedInlinkedin.com/company/titan-ml

Revenue estimate

Valuation estimate

Number of profiles

Funding overview

Funding rounds3 records

Investors4 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Doubleword