Pricing · Pay per token

Three tiers,
no contracts.

Priced like the API, billed like a membership. Per million reps (input tokens). No peak-hour surge, no annual lock-in. Cancel mid-set if you want.

Tier 01 · Open source
Self
host
$0.27
per 1M reps · input rate
What's included
  • DeepSeek V3 + Llama 405B
  • Run on your own hardware
  • Open weights, full fine-tune access
  • Data never leaves your gym
  • 128K context window
  • No SLA, no apology emails
  • Community support only
Spin up locally
Most chosen
Tier 02 · Mainstream
Stand
ard
$1.25
per 1M reps · input rate
What's included
  • GPT‑5 + Gemini 2.5 Pro
  • 400K – 2M token context window
  • Multimodal — video, voice, text
  • Real-time form analysis
  • Conversation history + memory
  • Output billed at $10 / 1M reps
  • Email support, 24h response
Start training
Tier 03 · Frontier
Elite
1:1
$15.00
per 1M reps · input rate
What's included
  • Claude Opus 4.1 + GPT‑5 Pro
  • Extended thinking · deep reasoning
  • Will refuse to let you ego-lift
  • 200K – 400K context window
  • Priority access, no rate limits
  • Output billed at $75 / 1M reps
  • Dedicated Slack support channel
Request Frontier

All six
trainers.

Every model on staff, their context window, and what they cost per million tokens. Mix and match within your tier at no extra charge.
Trainer
Context
In / 1M
Out / 1M
Tier
CL
Trainer 01 · Anthropic
Claude Sonnet 4.5
200K
$3.00
$15.00
Frontier
GP
Trainer 02 · OpenAI
GPT GPT-5
400K
$1.25
$10.00
Standard
GM
Trainer 03 · Google
Gemini 2.5 Pro
2M
$1.25
$10.00
Standard
GK
Trainer 04 · xAI
Grok Grok 4
256K
$3.00
$15.00
Frontier
LM
Trainer 05 · Meta
Llama 405B
128K
Free*
Free*
Self-host
DS
Trainer 06 · DeepSeek
DeepSeek V3
128K
$0.27
$1.10
Self-host

* Self-hosted models require your own infrastructure. Forge provides the tooling; compute costs are yours.

Common
questions.

What counts as a rep (token)?
One token ≈ 4 characters of text. A typical coaching session exchanges 2,000–8,000 tokens. At Standard tier, a full week of daily check-ins costs under $0.10.
Can I switch trainers mid-program?
Yes. All six models read from the same training log. Swap at any time — your history, PRs, and notes travel with you.
Is output really billed separately?
Yes, just like the underlying APIs. Input (your messages + context) and output (the coach's replies) have different per-token rates. We show you a running cost estimate in the session view.
Do I need a credit card to try it?
No. Sign up free and explore the dashboard. You only enter payment when you send your first message to the AI Coach. First 10K tokens are on us.
What is Self-host and do I need a server?
Self-host means you run DeepSeek V3 or Llama 405B on your own hardware. Forge gives you the integration layer; you supply the GPU. A capable local machine or a cloud VM works fine.
Is there a rate limit on Frontier?
No hard limits on Frontier. Claude Opus and GPT-5 Pro operate at priority queue level. Heavy users (>50M tokens/month) get a dedicated capacity allocation by default.
Ready when you are

Start
training.

First 10,000 tokens free. No credit card to sign up. Cancel anytime.

Start TrainingTalk to AI Coach first