The Grid is live: the first real-time spot market for AI
The Grid is a real-time spot market where suppliers compete to serve your AI requests, and you pay market price instead of list price.
If you're buying AI models today, you're almost certainly paying more than you need to.
Not because you're doing anything wrong. Because the system wasn't designed to show you what you're actually spending, or why. It simply does not allow price discovery.
Most teams default to the largest, latest model for everything. It's the path of least resistance. You're running Claude Opus 4.7 on customer support responses. GPT-5.5 on internal summaries. Gemini Pro on autocomplete suggestions. These are frontier models built for complex reasoning, and you're using them to say "your order has shipped." That's like chartering a private jet for a trip to the corner shop.
On top of that, the providers selling you AI models aren't optimising for your cost. They're optimising for their revenue. Models change constantly.
GPT-4 to GPT-4 Turbo to GPT-4o to GPT-5.5. Claude 4 to Claude 4.7. Each release is positioned as an upgrade, and each one nudges you toward a higher price point. Meanwhile the model you spent three months tuning your prompts around gets deprecated with a few weeks notice. There's no transparency into what you're paying relative to what the work actually costs. Just a bill at the end of the month that's higher than it should be.
This isn't a niche problem. It's structural. And it compounds as you scale.
Why this happens: AI models has never had a real market
Think about how oil gets priced. Or electricity. Or any commodity that matters at scale.
At some point, each of them went through the same transition: from fragmented, opaque, provider-controlled pricing to transparent, competitive, market-discovered pricing. Buyers got competition. Sellers got predictable demand. Prices reflected actual supply and demand rather than whoever set the list price.
AI models haven't had that moment yet.
Supply is fragmented across dozens of providers. Demand is unpredictable. There's no standard way to compare quality across providers. No competition at the point of purchase. No reference price anyone can trust. Just individual provider relationships and whatever rate you negotiated, or accepted.
The conditions that created commodity markets for oil and electricity exist right now for AI models. The infrastructure just hadn't been built yet.
What the Grid fixes:
The Grid is a real-time spot market where suppliers compete to serve your AI requests, and you pay market price instead of list price.
Instead of buying a specific model from a specific provider, you buy a quality tier, defined by objective benchmarks, not brand names. Multiple suppliers compete to serve your request at that quality level. The cheapest qualifying supplier wins. That competition is what drives your cost down, automatically, every time you make a call.
Three tiers at launch:
| Text Max | Frontier reasoning. Intelligence Index ≥53, 1M context, ≥30 tok/s throughput. For the tasks that need to think hard and get it right. |
| Text Prime | Production workloads. Intelligence Index ≥38, ≥128K context, ≥40 tok/s. For the everyday calls that need to ship reliably. |
| Text Standard | High volume. Intelligence Index ≥18, ≥100 tok/s, ≤1.32s TTFT. For batch jobs, autocomplete, the calls that just need to move fast. |
In preview (Code and Agent tiers):
Code Max, Code Prime, Code Standard, for coding workloads. From complex multi-file development and architectural decisions to fast autocomplete and batch edits.
Agent Max, Agent Prime, Agent Standard, for agentic workflows. From autonomous multi-step research and deep tool chains to high-throughput automation and orchestration loops.
Every tier has a quality floor enforced by continuous benchmark evaluation. Any model that meets the bar qualifies. Any model that drops below gets pulled automatically. The benchmark is the guarantee, not the brand name, not the provider's word.
How it works
The default experience is simple: you set a tier, make an API call, and The Grid routes your request to the best qualifying supplier at the best available price. You don't need to think about the mechanics. It works exactly like pay-as-you-go, you just get market price instead of list price.
One OpenAI-compatible API. Three lines of code to switch from your existing setup. No new SDK. No new dashboard unless you want one.
For teams who want more control, limit orders let you set the maximum price you're willing to pay. Set a limit for Text Prime at $0.50 or less, when the market clears at or below your price, your request fills. If it doesn't clear there, nothing happens and nothing gets charged. This is particularly useful for non-time-critical workloads: batch processing, scheduled jobs, anything where you'd rather wait for a better price than run immediately at a higher one.
Why this matters for you right now
AI costs are going up, not down. Models are getting more capable and more expensive. The gap between what you're paying today and what a competitive market would charge for the same output is real, and it widens as your usage scales.
The question worth asking isn't whether to optimise your AI spend. It's whether you want to spend engineering time doing it manually, prompt engineering, model selection, rate limit management, or whether you want a system that does it structurally.
The Grid is that system. Same quality. Lower cost. No manual optimisation required.
The right tier for each task, at the price the market sets. That's the shift we're offering.
We're in beta today
If you're a startup spending real money on consuming AI, we'd love you to try it.
thegrid.ai/docs to get started. Takes about 15 seconds to switch.
And if you have questions, thoughts, or feedback on what you'd like to see, reply here or find us at thegrid.ai. We're a small team and we read everything.
Your first 200M tokens are on us, start building
