TypeSafe Jev pricing (2026): $0.042 per million tokens, output free

The actual TypeSafe Jev pricing

Here is every published number in one place. TypeSafe does not have a dedicated pricing page, and every URL variant I tried (typesafe.ai/pricing, docs.typesafe.ai/pricing, typesafe.ai/docs/pricing) returns a 404. The figures below come straight from the homepage and the launch post.

The illustrative side-by-side on the homepage puts a single Jev call at $0.000081, versus $0.013880 for an LLM doing the same shaped work. That is the number TypeSafe wants you to screenshot, and for a pure decision call it is a fair one.

Why the output-free part matters

In most LLM pricing models, output is the expensive half, often around 5x the input rate. That is because a chatbot's value is the paragraph it writes, and paragraphs are a lot of tokens. Jev flips this. A System One model does not generate prose, it returns a typed decision plus a confidence score, so the output payload is a handful of bytes. Pricing it at zero costs TypeSafe almost nothing and removes the scariest line from your bill.

The practical read: if your workload really is a stream of small typed questions, "is_urgent", "which department", "how frustrated is this customer", then Jev's meter runs almost entirely on the input state you send, and that state is usually small. This is the same shape as the ticket triage work Cloudflare uses as its own Jev example, which is not a coincidence.

Model your own bill

Numbers on a homepage are one thing, your actual monthly volume is another. Plug in how many decisions you run per month and the average input size, and this compares Jev against a typical LLM at TypeSafe's own quoted range. The LLM rates are editable, because "typical" depends on which model you use.

A word of caution on reading that multiplier: it is a token-cost comparison for the decision step alone. It is not a total-cost-of-ownership number, because a real support flow does more than decide.

What the price does not include

This is where a pricing post earns its keep. The $0.042 rate is real and it is low, but it buys you a narrow thing. Here is what sits outside the meter:

• Generation. Jev cannot write the reply. It returns a decision, not a sentence. If your support workflow needs a drafted answer, you are still paying an LLM for that half.
• Orchestration. You have to build the loop that sends state in, reads the decision out, and acts on it. The token cost is trivial, the engineering is not.
• Actions. Deciding "escalate this" is free-ish. Actually escalating, tagging, refunding or routing the ticket is your integration work.
• A predictable seat or resolution price. There is no plan, no SLA tier, no committed-use discount published. For a team that needs to forecast a budget against human cost, "one input rate and a waitlist" is thin.

None of that makes Jev bad value. It makes it a component. The right mental model is that Jev prices the decision primitive, and the rest of your customer service automation budget lives elsewhere.

Is it subsidized? TypeSafe says: maybe

Cheap AI pricing always raises the "are they burning VC money" question, and TypeSafe answers it head-on. From the launch post:

"We make our pricing transparent. We can't prove it isn't subsidized; we'll need the long-term to prove the sustainability of our pricing (which we expect to go down, not up)."

That is more honest than most launch copy. The counter-evidence is technical: on Hacker News, one commenter reverse-engineered the price into a size estimate.

"At 0.042-per-million price they have, Astra estimates the model to be 3bn parameters."

paraschopra, Hacker News

A ~3-billion-parameter model is cheap to serve, so a low sustainable price is plausible rather than fantasy. The top comment on the thread still pushed back on the framing though, arguing the speed and cost comparisons flatter Jev because it skips generation entirely:

"Seems like a more accurate title would be 'Jev: Trading general purpose generation for fast typed inference'... the speed comparison seems misleading?"

jacobgold, Hacker News

The pricing enthusiasm on X was real too, and from a named operator. Vercel's CEO reported it beating GPT Luna on a production reviewer:

"Jev is up to 18x faster (p95) and more accurate. It's coming to @vercel AI Gateway and likely new default."

Guillermo Rauch (@rauchg), X

Read those together and the pricing verdict is: the rate is real and low, the "238x" is a best-case single-call comparison, and the honest use is decisions, not everything.

Where this leaves a support team

If you run a helpdesk, the interesting question is not "how cheap are the tokens", it is "what does the finished automation cost and do". A tier-1 deflection flow that triages, drafts, and resolves needs generation and actions on top of the decision. Jev can be the fast, cheap triage brain inside that, but it is not the AI agent itself.

This is the split I keep coming back to: Jev is infrastructure, a decision primitive you build with. eesel is the employee. Our AI helpdesk teammate joins your existing queue, makes the triage-route-escalate call and writes the reply and takes the action, then simulates the whole thing against your historical tickets before it goes live. We price per resolution, not per token, so the number on the invoice maps to work actually done, not to how many yes/no calls happened under the hood. It is one of a small roster of ready-to-work AI teammates, not a raw model you assemble yourself.

And if you live in a terminal, the same teammate is drivable from the eesel CLI. It is not a second product bolted on, it is the same eesel workspace and agent exposed for scripting: a person can run it by hand, a cron job can automate it, and a coding agent like Claude Code or Cursor can drive it through the CLI and eesel's MCP server. So if the reason you were pricing Jev was to wire a decision model into an agentic pipeline from the command line, you can point that same pipeline at a teammate that already knows your help center and can act on the ticket, not just score it.

Want to see the finished cost, not the token cost? eesel runs on your own tickets in a free trial, so you can watch what a resolution actually costs before you commit.

The bottom line on TypeSafe Jev pricing

Jev is $0.042 per million input tokens with output free, and for a pure decision workload that is a very low number. It is bounded by a real caveat: no plan tiers, no public free tier, a waitlist for access, and a price the company admits it cannot yet prove is sustainable. Model it as a cheap decision primitive, not as a drop-in replacement for the LLM writing your answers, and the TypeSafe Jev pricing story is easy to reason about. Whether you should build your whole support stack on it is a different question, and one worth answering with the full review open next to this.