Modelglass Review: A Free AI Model Pricing Registry With BYOK Routing Built In

Keeping up with AI model pricing has become its own part-time job: new models ship weekly, providers change prices without much notice, and benchmark numbers get quoted without saying how they were measured. Modelglass is a sourced, daily-updated registry that tracks what every model actually costs and what it is good at, then goes a step further with a VS Code extension and MCP server that can route a coding task to the cheapest model that still clears a quality bar, using your own provider key, with no proxy in between.

What Modelglass Does

Modelglass is a pricing and capability registry for AI models. It sits in front of your prompt, not behind it: read the data yourself to decide which model fits a task, or wire it into your own workflow and let it choose for you. At the time of writing it tracks 174 models across 56 providers and five model types (language, image, video, audio, and 3D), with every price independently sourced against each provider's own pages and refreshed daily.

The comparison layer lives at the registry itself, where you can browse by modality (46 language models, 21 image, 23 video, 21 audio, 8 3D at last count), filter by provider, sort by price, and run a "Compare Two Models" diff that reports the unit-matched price delta between two models, whether the price has been stable over time, a per-dimension capability difference, and any billing-unit changes to watch for. The compare tool works across every modality on every plan, including the free tier.

Who It Is Built For

Developers and AI engineers comparing API pricing, context windows, and provider availability before committing to a model in a product. The registry's structured metadata is built to be read programmatically through the API and MCP server, not just browsed.

Teams evaluating a model or provider switch who want to know the real cost and capability gap before making the call, rather than discovering it after integrating. The Compare Two Models tool is built specifically for this: a grounded migration diff instead of a guess.

VS Code users, through the Modelglass Cost-Aware Router extension, who want routing decisions made for them inside the editor. Route Task recommends the cheapest model that clears a benchmark bar for a coding or writing task, at no cost and with no provider key required. Run Task goes further and actually executes the call, using your own configured provider key(s), ranking against live SWE-bench Verified, Aider Polyglot, and similar benchmark scores depending on the task category.

How It Works

On the web, the flow is straightforward: open the registry, filter to the modality you care about, and compare price per unit (per image, per megapixel, per token, and so on) alongside architecture and availability status. The "Estimate monthly cost" tool projects what a given usage pattern would actually cost across models before you commit.

Inside VS Code, the free Route Task command infers your task type from the active file and recommends a model without ever calling a provider API. The paid-tier-gated Run Task command goes one step further: it executes the call against your own key, with no Modelglass proxy in the request path at any point, and automatically falls back to the next-best model on a different provider if the first call fails, when you have more than one provider key configured. The same router is exposed inside GitHub Copilot Chat as a selectable model, and separately handles BYOK video generation through Runway and audio generation through ElevenLabs.

Features

• Daily-refreshed pricing and capability data across five modalities, with every figure sourced and dated rather than guessed when it cannot be verified.
• Compare Two Models, a grounded price and capability diff with unit-matched deltas, price-stability history, and billing-unit change warnings, available on every plan including Free.
• API and MCP server access for querying pricing and capability data programmatically, so the registry can back a routing layer, a pricing page, or an agent's own decision-making.
• A VS Code extension with two distinct commands, Route Task (recommendation only, no key needed) and Run Task (BYOK execution with automatic provider fallback on the Pro tier), plus BYOK video and audio generation through Runway and ElevenLabs.
• A free iOS app for quick pricing lookups on the move, no signup required.
• No proxy in the request path. Modelglass states plainly that it does not see your prompts or your outputs. It is a registry you or your tools query, not a service that sits between you and the model provider.

Pricing

The open comparison registry is free to browse with no account needed. Modelglass's own tiers apply to the API, MCP access limits, and VS Code extension capabilities: Free ($0) includes current pricing data, MCP access, all API endpoints, 100 requests per day, and the VS Code extension with unlimited comparisons using one provider key and no retry. Starter ($9/month after a 7-day free trial) adds 12 months of pricing history and 10,000 requests per day. Pro ($39/month after a 7-day free trial) adds full pricing history, 100,000 requests per day, support for two or more provider keys with automatic fallback retry across providers, custom routing rules, and a 99.9% uptime SLA. Cancel before the trial ends on either paid tier and you are not charged.

How It Compares

The default way most people track model pricing is a bookmark folder of provider pricing pages, checked manually whenever something feels off, with no consistent unit to compare against and no record of what a price used to be. General-purpose "list of AI models" sites are often little more than a directory, without daily price verification or a capability benchmark tied to each entry. LLM routing proxies solve the routing problem but usually do it by sitting in the request path themselves, which means your traffic (and sometimes your provider keys) passes through a third party. Modelglass's router is explicitly BYOK with no proxy step, which is a meaningfully different trust model if you are routing production traffic rather than just comparing prices casually.

Bottom Line

If you are choosing between AI models on cost, or maintaining a product that depends on knowing when a cheaper or better option becomes available, Modelglass replaces a manual, error-prone comparison process with a sourced, daily-updated registry, and gives you a way to act on it directly in VS Code without handing your provider keys or your prompts to a middleman. The free tier alone covers most casual comparison use; the paid tiers earn their keep once you want price history or are routing real production traffic.

Frequently Asked Questions

Is Modelglass free to use?

Yes. The comparison registry itself is free to browse with no account required, and the Free API tier includes current pricing data, MCP access, all API endpoints, and the VS Code extension with one provider key. Paid tiers add pricing history and higher request limits.

Does Modelglass see my prompts or outputs?

No. Modelglass describes itself as a registry you read or query, not a proxy in the request path. The VS Code extension's Run Task command and its Runway/ElevenLabs generation commands execute directly against your own configured provider key, with no Modelglass server in between.

How many models does Modelglass track, and how current is the data?

At the time of writing, 174 models across 56 providers and five modalities (language, image, video, audio, 3D), refreshed daily. Every price is checked against the provider's own published pricing, and figures that cannot be verified are marked rather than estimated.

What does the VS Code extension actually do?

It offers two main commands: Route Task, which recommends the cheapest model that clears a benchmark bar for your task with no execution and no provider key required, and Run Task, which executes the call against your own configured provider key(s), with automatic fallback to another provider on failure if you have more than one key configured (Pro tier).

Do I need to know which AI model to use before trying Modelglass?

No. That is the problem it is built to solve. Browse the registry by modality and price to compare options directly, or let the VS Code extension's Route Task command recommend one for a specific coding or writing task based on live benchmark and pricing data.