grynx Join the waitlist

For heavy Claude Code & Codex users

2–3× the agent work. Same budget.

Grynx smart-routes every step of your agent session: open models like GLM, Qwen and DeepSeek handle the grunt work at open-model prices — frontier models step in only where they earn it. Keep your Claude Code or Codex workflow — one subscription, no API keys, no weekly walls.

Join the waitlist See how it works

Private beta · No credit card · Founding pricing locked for everyone who joins during beta

Smart routing across opus-5 sonnet-5 gpt-5.6-sol glm-5.3 qwen3.8 deepseek-v4 kimi-k3 picked per step, automatically — new releases join the pool within days, or pin any model yourself
August so far: $23.80 spent · $86 all-frontier equiv · 3.6× further

Illustrative session — the router picks the mix per task. Every session gets this receipt.

The problem

You're paying frontier prices for grunt work

Agent sessions are mostly reading files, running tests, and writing boilerplate. A fraction of the work needs a frontier model — but that's the rate you pay for all of it.

rate_limit_reached

The cap lands mid-sprint

Weekly limits reset on their schedule, not yours. So you ration — save the good model for "worthy" tasks and time your week around a quota. That's you doing a router's job, by hand.

$14.60 / task

Every step billed like it's hard

Scanning a repo costs the same per token as solving a race condition. Direct subscriptions can't tell the difference — so boilerplate burns the same budget as the breakthrough.

5 keys · 3 dashboards

A drawer full of keys

A Claude plan here, a ChatGPT plan there, API keys for everything else. Two tools that don't share history, and no idea what a task actually cost until the invoices land.

Why Grynx

One subscription. Every engine. Audited routing.

Your workflow, no forks

Grynx launches the official Claude Code and Codex CLIs through their supported gateway hooks — same TUI, same muscle memory, no Anthropic or OpenAI account required — or runs its own native engine. One history, one bill, switch per task.

Routing you can audit

Each step goes to the cheapest model that passes our evals for that step type — hard reasoning still goes frontier. Every task gets a receipt: what ran where, what it cost, what all-frontier would have cost. Pin any session to any model whenever you want the control.

Capacity that doesn't cliff

No weekly wall. You set the caps — per task, per day, per month — enforced while agents run, so a runaway loop stops at your number instead of on your invoice.

How it works

Same terminal. New engine room.

Install & sign in

npm i -g grynx, sign in, done. No API keys to create — usage is included in your plan.

Launch your engine

grynx claude or grynx codex. The exact CLI you already use — Grynx sits at the model layer, not in your prompts.

Every call, routed

Simple steps go to fast open models, hard reasoning goes frontier. Evals decide the cut per step type — quality is gated, not gambled.

Read the receipt

Per-task cost, model mix, and savings vs all-frontier — live on your dashboard. Set caps once, stop thinking about the invoice.

Founding pricing

Pick your capacity

Join during beta and these prices are locked for your account. Usage beyond your plan bills at the same routed rates — under hard caps you control.

BYOK

$0 / month

Bring your own API keys

  • Claude Code & Codex, one workspace
  • Unified sessions & history
  • Smart routing across your keys
  • Cost receipts on every task
  • You pay providers directly
Join the waitlist
Most popular

Pro

$49 / month

≈ $120–150 of all-frontier work included

  • No keys — managed usage included
  • Full smart routing, eval-gated
  • Pin any task to any model
  • Hard caps: per task / day / month
  • Overage at the same routed rates
Join the waitlist

Max

$149 / month

≈ $400–500 of all-frontier work included

  • Everything in Pro
  • Higher concurrency — parallel agents
  • Priority capacity at peak hours
  • Early access to new engines & models
Join the waitlist

Agency or team? Pooled budgets, per-member caps, one invoice — mention it in the form and we'll reach out.

FAQ

Fair questions

Won't cheaper models write worse code?

For the steps they're given — no, and that's the whole gate. We continuously evaluate engine × model × step-type combinations, and a model only enters the routing pool for step types where it matches frontier acceptance in our evals. Hard reasoning and debugging still go to frontier models. And you can pin any session or step to any model, any time.

Do I need my own API keys, or a Claude / ChatGPT subscription?

No. On Pro and Max the CLIs authenticate against the Grynx gateway and usage comes out of your plan — you don't need an Anthropic or OpenAI account at all. BYOK stays available on every plan if you'd rather pay providers directly; Grynx is then your workspace, router, and receipt layer on top of your own keys.

Is this a fork of Claude Code or Codex?

No. Grynx launches the official CLIs, unmodified, using their built-in gateway configuration, and provides the endpoint their model calls route through. Your workflows, config, and muscle memory carry over. Grynx also ships its own native engine, so you're never locked to either CLI. Grynx is an independent product with no affiliation to Anthropic or OpenAI.

How can it be this much cheaper?

Two reasons. We run our own model-routing infrastructure and buy capacity at scale, and — more importantly — routing means frontier tokens are only spent where they earn it. Most agent work doesn't need them. Most of that efficiency is passed on to you; that's the product.

When do I get access?

We onboard in weekly waves, in waitlist order. Everyone who joins during beta keeps founding pricing for the life of their account.

Early access

Get off the ration

We onboard a small wave each week. Tell us what you run today and we'll size your capacity.

If Grynx worked as described, what would it be worth to you per month?

Private beta · No credit card · Founding pricing locked at launch