When Giants Fight, Does the One in the Middle Win by Not Taking Sides?

Deep thoughts on AI and aspirations —— ByteDance Deep Thinking Circle

In January, Anthropic did something that seemed self-serving: they blocked OpenCode. Any request with “open code” in the system prompt was rejected outright, effectively cutting off users who accessed Claude subscriptions through OpenCode.

The move backfired spectacularly, thrusting OpenCode into the spotlight. The block inadvertently elevated it to the same tier as Claude Code—even people who’d never used it took notice that Claude Code felt threatened enough to act against it. Days later, OpenAI announced official support for OpenCode. Users flooded in.

According to the OpenCode team’s own interviews, the product went from zero to roughly 13 million monthly active users in a year, processing about 7 trillion tokens daily, with annualized revenue approaching $60 million. For reference, they mentioned OpenRouter’s entire platform processes around 6 trillion tokens daily. Take these numbers with a grain of salt since they come from the company itself, but the growth trajectory is striking nonetheless.

In an era where everyone says “the model is the product,” every model provider has launched their own coding tool, and star products like Cursor are already out there. How does an open-source programming agent emerge from the gap between Claude Code and Codex?

Winning Position, Not Product

The obvious explanation is superior product quality. But OpenCode co-founder Dax Raad’s answer is the opposite, stated quite bluntly: most of their success came from getting the positioning right. He even said that if product quality were half of what it is now, the outcome would probably be similar.

That stings, but it’s honest. Functionally, what they built isn’t fundamentally different from Claude Code or Codex—they’re all agents that help you write code. What’s truly different is the position they chose to occupy: the open-source position.

Their bet was that while coding agents were everywhere, no one had claimed “we’re the open-source option,” and that territory was valuable. The reasoning was simple: historically, almost all developer tools eventually go open source. Databases are open source, compilers went from proprietary to open source—look at any tech stack and the major components are open source.

So at launch they claimed support for over 70 models and providers, and built an open-source database called models.Dev cataloging various models and providers. These moves had one clear purpose: claim that position and attract every user who doesn’t want to be locked into a single model. Even some well-funded large companies use them, citing the desire to avoid lock-in to any particular model or framework.

Why Giants Can’t Crush It

Why is this “neutral” position so valuable? First, look at today’s model competition landscape.

OpenCode’s core thesis: competition between models will only intensify, closed-source models compete with each other while open-source models chase them, and prices face constant downward pressure. We’ll test this thesis shortly. First, the inference.

For anyone using models to get work done, the biggest risk is betting on the wrong horse and getting locked in. The market needs an intermediary layer that doesn’t bind to any model and allows switching. And this intermediary layer is useful to every model provider. It gradually becomes industry infrastructure—no giant would deliberately destroy it because doing so wouldn’t benefit them either.

Competition doesn’t hurt it; it amplifies it. Revisit the blocking incident: after the block triggered user outrage, OpenCode immediately messaged OpenAI saying tomorrow everyone will be upset with Anthropic, you can take the opposite stance, officially support OpenCode, and win the PR battle. OpenAI confirmed the next day and announced official support that same day.

The team distilled this playbook into one sentence: choose a temporary “villain,” unite all competitors against it. The giants each have their own calculations, but as long as they compete, they’ll race to endorse the neutral party. This is leverage even small companies can use.

This position has another benefit: it makes trying new models cheap. When a new model launches, users can switch and test directly in this neutral tool, switching back anytime if unsatisfied. Providers wanting to win users with new models end up depending on this intermediary layer to reach them. The fiercer the competition, the more comfortable this position becomes. In the founder’s words, whenever one model tries to beat another, they benefit.

The Hidden Premise

Don’t rush to treat this playbook as a universal formula. This position’s viability hides a premise: competition must persist.

If the model market becomes winner-take-all one day, with only one dominant player, or if several providers band together to close their APIs and only build their own applications, then user demand for “switchability” disappears, and the neutral layer becomes useless.

This isn’t fear-mongering. OpenCode’s co-founder acknowledged the risk of this path being cut off: structural tension exists within labs, where product teams want to lock users in with model-product binding, while sales teams must open APIs to hit revenue targets. As long as this internal struggle persists, providers struggle to go fully closed. But “struggle” doesn’t mean “won’t.” If one lab someday puts revenue targets aside and commits to locking down the ecosystem, this neutral position gets pulled out from under.

In other words, the entire value of the neutral position depends on a variable it cannot control: multipolar competition must continue. Once this premise fails, the position rapidly loses value.

Surviving Is One Thing, Making Money Is Another

Traffic alone isn’t enough—products in the gaps must monetize. OpenCode’s approach is worth unpacking.

They don’t monetize through ads or tool subscriptions, but treat free tiers as customer acquisition cost. The founder’s view: the free tier is the new CAC, tokens are advertising. Users experience the value of programming agents through free credits before they’re willing to pay.

Actual revenue comes from two streams. One is inference aggregation: many users can’t handle multiple model accounts, rate limits, and integrations themselves, so OpenCode provides a unified entry point, aggregating multiple inference services for a fee. The team disclosed this business, launched about eight months prior, reached nearly $40 million annualized with healthy margins. The other is enterprise control panels: companies with thousands of engineers can’t have everyone configure their own keys—they need unified management of models, permissions, and budgets.

Notice both revenue streams grow from the same position: because you’re neutral and don’t lock in, users dare to hand you the entry point; because you aggregate all models, you’re entitled to charge this “access fee.” Position isn’t just a moat—it’s the reason for charging. This is more concrete than simply “build goodwill with open-source freemium.”

Another Kind of Moat

This offers a different insight.

Most people understand moats as technology, product, brand, network effects—the accumulated advantage playbook. But the OpenCode case demonstrates another kind: the positional moat. Its value comes from occupying an ecological niche, not from being better than others.

Positional moats fail differently than technical moats. Technical moats fail when surpassed; positional moats fail when the battlefield itself changes. The former is visible—just chase it. The latter is hard to defend because your value comes from others’ conflict, not your own accumulation.

So I’d rather read this as a conditional pattern, not preach it as a success formula. When choosing direction, don’t rush to compare technical strength—first ask: will this market remain multipolar long-term? If multipolar, the neutral position is worth claiming; if destined for monopoly, the neutral position is a trap.

When looking at any project labeled “open source, neutral, compatible with all models,” ask one more question: what premise does its value rest on? Does that premise still hold?

The fiercer giants fight, the more valuable the middle ground. That statement is only half right. The complete version: only if they keep fighting does this ground stay valuable.