Pro Reasoning
Frontier chain-of-thought models — DeepSeek R1-0528, Qwen3 235B, Llama 3.3 70B. 128K context, transparent per-token billing.
Token Factory · Frontier Inference Platform
Frontier inference,
Frontier inference,
one endpoint.
Token Factory is the inference layer for modern AI — 60+ open-source and frontier models through one OpenAI- and Anthropic-compatible API, with transparent per-token pricing and zero infrastructure to manage.
/ The Model Matrix
Four flagship surfaces.
Four flagship surfaces.
One API underneath.
UltraFast LPU
Latency-critical inference on Language Processing Unit silicon. Up to 1,000 TPS for real-time voice, agents and interactive tools.
Music 2.6
Speech 2.8
Image Studio
Vector Engine
1,000,000Free tokens on signup
$3Starter plan / 24 hours
0Credit card required
60+Models unlocked
Get your Opus Million tokens today.
Your million tokens from $3, same plan.
Serving models from
DeepSeek
Mera
Meta Llama
Qwen
Mistral
OpenAI
Anthropic
OpenAI OSS
Black Forest
Moonshot
Stability
DeepSeek
Mera
Meta Llama
Qwen
Mistral
OpenAI
Anthropic
OpenAI OSS
Black Forest
Moonshot
Stability
Mera
Meta Llama
Qwen
Mistral
OpenAI
Anthropic
OpenAI OSS
Black Forest
Moonshot
Stability
DeepSeek
Mera
Meta Llama
Qwen
Mistral
OpenAI
Anthropic
OpenAI OSS
Black Forest
Moonshot
Stability
/ The Platform
Built for production
Built for production
from the first request.
Unified API
One endpoint. Every modality.
Text, reasoning, code, vision, speech, embeddings and image generation — all through one OpenAI- and Anthropic-compatible interface. Swap providers by changing one URL.
# OpenAI & Anthropic compatible, drop-in
from openai import OpenAI
client = OpenAI(
base_url=“https://api.tokenfactory.io/v1/”,
api_key=“tf-•••••”)
Throughput
1,000TPS
Peak tokens per second on GPT OSS 20B UltraFast
Latency
<100ms
Time-to-first-token under load
Audio
Music 2.6 — generative audio at studio quality
Lyrics, stems, transcription and voice conversion in one API.
Pricing
From $3. No idle GPU cost.
1M-token test plan, monthly windows, or pure pay-as-you-go.
SLA
99.9%
Uptime backed by multi-region failover
Caching
50% off cached input
On GPT OSS models — automatic prompt caching.
Compat
Drop-in OpenAI & Anthropic
Works with every OpenAI and Anthropic SDK out of the box.
/ Why Token Factory
One platform. Every workload. Zero lock-in.
From your first prototype to multi-region production scale, the same API, the same billing model, the same observability stack.
Performance
UltraFast LPU inference
Up to 1,000 tokens per second on Language Processing Unit silicon. Sub-100ms time-to-first-token, even at peak load.
Compatibility
OpenAI & Anthropic compatible, drop-in
Works with every OpenAI SDK, every Anthropic SDK, the Vercel AI SDK, LangChain, LlamaIndex, the OpenAI CLI, and Claude Code. Change one base URL — keep the rest.
Pricing
Transparent per-token billing
No idle GPU cost, no minimums, no surprise overage. Stripe billing, Odoo finance sync, line-item invoices for every request.
Catalog
60+ models, one API
Reasoning, code, multilingual, speech, vision, embeddings, image generation. Frontier and open-source, all under one key.
Reliability
99.9% SLA, multi-region
Automatic failover across three regions, real-time health monitoring, and a public status page. Your apps stay up.
Developer experience
First-class SDKs
Python, TypeScript, Go, and Rust SDKs. Type hints, streaming, retries, and full observability baked in. Five lines to your first request.
/ By the numbers
Built for teams that ship.
12B+
Tokens served daily
60+
Models in catalog
99.97%
Uptime last 90 days
14K+
Active developers
/ Latest Updates
Shipped this month.
Model · Reasoning
NEW
NEW
DeepSeek R1-0528 — chain-of-thought at frontier parity
128K context, $0.55 / $2.19 per M tokens. Now serving in production with stable latency.
Infra · UltraFast
NEW
NEW
LPU silicon — 1,000 TPS on GPT OSS 20B
8 UltraFast models now live on Language Processing Unit silicon. Sub-100ms TTFT under load.
Pricing
UPDATE
UPDATE
Token Plan — $3 test plan with 1M tokens
Try every model with no card. Predictable monthly windows from $25. Pure PAYG for scale.
Get $5 in free credits.
Ship your first agent today.
Sign up, get an API key, and start calling frontier models in under 60 seconds. No card required.
/ One-Line Setup
Set up your IDE in 30 seconds.
Run one command. Paste your API key. Token Factory auto-configures Claude Code, Codex, and every model in your workspace based on the plan you’re on — no env files, no boilerplate.
✓
Auto-detects your IDE — Claude Code, Cursor, Codex CLI, Continue, Cline, or any OpenAI/Anthropic-compatible client.
✓
Plan-aware model routing — picks the right default model, context window, and rate limit for your tier (Test, Starter, Max, Enterprise).
✓
Writes the env file for you —
.env.local with OPENAI_BASE_URL, ANTHROPIC_BASE_URL, and TF_API_KEY in one shot.
✓
Safe to re-run — upgrades are idempotent. Existing keys, custom models, and overrides are preserved.
CLI · v1.2.0
terminal
Install with npx
Works on macOS, Linux, and Windows. Node 18+ required.
terminal
# Install the CLI & set up your workspace
$ npx token-factory init
$ npx token-factory init
# Paste your API key when prompted
? Enter your Token Factory API key: tf-prod-****-****
# Pick your IDE
? Detected: Claude Code · Cursor · Codex · Continue
› All of the above
# Models are routed by your plan
✓ Plan: Token Max
✓ Default model: claude-3-5-sonnet-ultra (UltraFast)
✓ Context window: 128K
✓ Wrote .env.local · .claude/settings.json · .codex/config.toml
✓ You’re ready. Run claude or codex to start.
~ 1.4 MB · MIT licensed
Copy
✓ Test
✓ Starter
✓ Max
✓ Enterprise
✓ Starter
✓ Max
✓ Enterprise