Lightcone

Jay V on Why OpenCode Bets the Whole Field, Not One Model

Jay V· Founder and CEO of OpenCode at OpenCode
·~44 min·English·Y Combinator
AgentsOpen SourceInferenceBusiness Strategy
TL;DR

OpenCode's Jay V explains how a model-neutral, open-source coding agent reached ~13M users and ~7T tokens a day by betting the entire open-weight field and serving the world the frontier labs price out.

01The Scale

The 20x Year

OpenCode ended June at roughly 13 million monthly users and now moves about 7 trillion tokens a day — more than OpenRouter's entire total, up from around 300 billion in January.

We also recently started processing around 7 trillion tokens per day.

— Jay V, Lightcone
Key Insight
A single third-party harness now out-volumes OpenRouter's aggregate — the token layer, not the model layer, is where usage concentrates. When one client is the biggest buyer for most open-weight labs, it stops being a customer and starts being distribution.
02The Inflection

The Ban That Backfired

When Anthropic tried to block Claude Code subscriptions inside OpenCode by rejecting any request whose system prompt said "open code," it accidentally put the two products on the same pedestal and sent OpenCode's users climbing.

what it inadvertently did was it put open code and Claude code on the same sort of pedestal. It like equated the two products in some ways

— Jay V, Lightcone
Key Insight
The block was economically rational — Anthropic was subsidizing tokens it didn't want spent elsewhere — but it leaked a signal: the incumbent treated OpenCode as a peer. In a crowded field of a dozen coding agents, being singled out is the cheapest possible endorsement.
03The Strategy

Betting the Field, Not the Winner

OpenCode refuses to pick a winning lab; it bets that the whole field of open-weight models — DeepSeek, GLM, Kimi, MiniMax — keeps getting better, and its own growth tracks each bump in that market.

it's not that we're picking a winner in terms of a model lab. We're just betting the field. We just think the rest of the field is going to do well.

— Jay V, Lightcone
Key Insight
This is a marketplace bet, not a model bet. If models are commoditizing into specialists — DeepSeek owning cost, others owning frontend or speed — then the durable position is the neutral layer that lets a user swap between them. Owning the choice is worth more than owning any one model.
04The Global Market

Token-Maxing vs. Token-Budgeting

The Valley throws money at tokens; most of the world budgets them — switching to ultra-cheap models to stretch a coding agent — which is why OpenCode's biggest single market is China, with heavy usage across Indonesia, Brazil, and Vietnam.

It's huge in developing countries like Indonesia is 4% of your traffic, Brazil is 5% of your traffic, places like Vietnam

— Lightcone, Y Combinator
Key Insight
The Valley optimizes for capability at any price; the rest of the world optimizes for access. Cheap open-weight models don't just win on cost — they open a market the frontier labs' pricing structurally excludes. OpenCode's geography is a map of who gets priced out.
05The Unit Economics

The Subsidy Chasm

Frontier labs pay users' way across an expensive learning curve; OpenCode instead uses a free tier for the first "aha" and a cheap subscription for real work, then turns its token volume into a discount that becomes margin as heavy users pay per token.

we're able to get volume discounts because of the amount of tokens that we sort of serve. And so, when you pay per token, that effectively turns into our margin.

— Jay V, Lightcone
Key Insight
Both models front the cost of the expensive expert-user chasm — but the incumbent eats it as a subsidy while OpenCode converts scale into a purchasing discount. Serving tokens still costs money, yet a per-token whale widens the margin rather than deepening a subsidy — which is how the inference business approached a ~$40M run-rate in about eight months without a war chest.
06The Infra Edge

The 24-Hour GPU

Because OpenCode's users are spread across the globe, demand from the East fills the trough while the West sleeps and vice versa, flattening GPU utilization into a stable 24-hour cycle that regional competitors can't match.

we have a reasonably stable 24-hour GPU cycle, which allows for pretty good utilization.

— Jay V, Lightcone
Key Insight
Global reach isn't just a growth story — it's an infrastructure hedge. A regional service pays for peak capacity it idles half the day; OpenCode's follow-the-sun demand keeps rented GPUs busy around the clock, turning geography into a structural cost advantage on the exact resource its whole business runs on.
07The Backstory

16 Years to an Overnight Success

The company behind OpenCode is a 16-year-old legal entity that filed nine YC applications before getting in, shipped a coffee-over-SSH terminal store, and only found runaway product-market fit when years of terminal-UI taste met the coding-agent moment.

It was really more a journey that took 10 years to to get to 0 to 30 million in 8 months.

— Lightcone, Y Combinator
Key Insight
The overnight-success framing hides the real lesson: none of the dead ends were wasted. A consumer company taught acquisition and metrics; the open-source projects and terminal-UI obsession built the exact taste that made OpenCode feel right on day one. The grind wasn't the cost of the win — it was the preparation for it.