New models land on OpenRouter almost every day. Tracking what dropped is easy. Knowing which one to actually use is harder.
DEV·TV already has a NEW MODELS channel for the first question. Today, I added channel 13: MODEL FACE-OFF, for the second.
Every bout is decided on specs and price, not intelligence. A slide says "wins on specs," never "better model." OpenRouter's catalog has no benchmark data, and I'd rather the channel be honest than impressive.
How it works
Every model added to OpenRouter this week is grouped by type (🧠 Reasoning, 💻 Coding, 👁️ Vision, 💬 Chat, 🎨 Image Gen, and more), then by price tier: Free, Budget, Mid, and Premium.
Models are paired with their closest rivals within the same division. A free reasoning model only fights another free reasoning model.
Each slide is a tale of the tape:
- Input price and output price: the cheaper model wins.
- Context window: the larger wins.
- Maximum output: the larger wins.
- Input modalities: more supported input types wins.
Reasoning isn't a row because it decides the division: every model in a reasoning bout supports it.
Each row is worth a point. If both models are level on a row, like two free models on price, nobody scores it. A missing stat skips the row instead of counting as a loss.
This morning's headline bout
MiniMax M3.1 Flash vs. Xing4.0 29B A4B, free reasoning division.
MiniMax M3.1 Flash wins 3–0 on specs.
Both models are free, but MiniMax brings a 1M-token context window, a 524K maximum output, and support for text, image, and video inputs. Xing4.0 29B A4B offers a 262K context window and 236K maximum output, with text input.
That's the kind of difference you might miss when browsing a catalog of model names and numbers.
The weekly leaderboard
The channel ends with a weekly leaderboard ranking new models by context per dollar. It also highlights the cheapest model and the model with the biggest context window.
The goal is to turn a constantly changing catalog into something you can watch, compare, and understand at a glance.
What it doesn't do
A bigger context window doesn't automatically mean better answers. A lower price doesn't guarantee better results for your workload.
MODEL FACE-OFF compares specifications and prices, not model intelligence. Quality data would be a valuable next addition, but until then, the channel sticks to what it can actually measure.
Still one file
Like everything in DEV·TV, MODEL FACE-OFF runs in your browser from a single index.html file.
- No API key
- No backend
- No login
- No setup or build step
It fetches data from OpenRouter's public models endpoint and performs the comparisons in your browser.
DEV·TV is my attempt to build a TV for the developer internet: something you can leave running in the background instead of constantly searching and scrolling.
MODEL FACE-OFF is another channel in that experiment.
📺 Try it: https://shouvik12.github.io/devtv/, then switch to channel 13.
⭐ Star it: https://github.com/shouvik12/devtv
What would you add to the scoring system? Price and specs are easy to compare. Measuring actual model quality is the harder problem.
Top comments (0)