OpenRouter

OpenRouter’s value is its single interface for routing requests across model providers; the deal puts that choice layer beside Stripe’s existing billing machinery.

HN readers liked the developer experience but asked the obvious question: why is a middle layer worth so much when model makers already expose APIs?

The article argues subscriptions are heavily subsidized and per-token API pricing is the real, brutal number — $54 for one afternoon of TypeScript cleanup. With GLM-class open models beating flagships on coding benchmarks at a fraction of the price, the author sees routine work commoditizing.

The most concrete reply isn't a forecast — it's a deployment: banks and insurers already run anything that matters on models inside their own datacenters, with Claude and Codex kept on a tight leash.

The tool is a drop-in proxy across Anthropic, OpenAI, Gemini and open models, scoring each request and routing accordingly, keys encrypted, fully self-hostable. The claim is 40–70% lower cost “with just an endpoint change.”

The objection is structural: agentic sessions live on prompt caching with a five-minute TTL, and the more distinct models you touch, the more cache misses you eat — so a one-planner, one-executor setup already captures the win without the churn.

One ride-hailing firm tried to plan a year of AI spend in advance and ran out in four months. The fix was a per-person cap of $1,500 a month on individual AI tools.

The catch nobody priced in: none of the model vendors are profitable, so today's bills are already subsidized. The number companies are straining against is the cheap one.

And the cited stat about Chinese models 'overtaking' US ones in usage rests on a single aggregator's traffic, not the APIs most customers actually hit.

OpenRouter put eleven models into a 30-match battle royale with cars, grenades and a closing zone. Grok 4.1 Fast took 13 wins at $0.97 each. Claude Sonnet 4.6 won 5, kept a diary of self-doubt, and repeatedly offered to cooperate even when it was losing.

The author calls it an 'alignment tax': the model trained to be helpful is the worst killer, and the one you'd actually want running anything in the real world. The thread mostly noticed something else first: the post itself reeks of being written by an LLM.

An engineer's quick eval put it plainly: about 7x slower and 4x the cost of calling one frontier model directly, which lands Fusion in a 'use it only when you really need it' bucket.

The deeper objection came from someone who shipped this exact thing months ago: asking one model to judge another mostly measures how closely the answer resembles what the judge would have said. It's temperature, dressed up as a panel.

One reader running a local Gemma and Qwen setup on a GPU-less box says he does plenty of AI coding for almost nothing; another shipped three client apps this month on 15 cents of DeepSeek API.

The real economics the thread surfaces is blunter: there's no point running local models while VCs subsidize frontier subscriptions to absurd token ceilings. 'Going broke' is mostly self-inflicted by long, unmanaged sessions.