Breaking

OpenRouter MCP Adds Design Arena Benchmarks for Live Model Selection in Agents

June 25, 2026 at 14:02 EDT

AI Agents
Foundation Models
Software Dev & Coding

On June 25, 2026, OpenRouter announced that it had integrated Design Arena's benchmarks into "OpenRouter MCP," enabling AI agents to reference real-time model performance and pricing information. Agents can now select and compare models based on live scores rather than guesses derived from stale training data.

June 25, 2026 · OpenRouter

Agents Can Now Pick Their Own Model — In Real Time

The new OpenRouter MCP Server lets AI agents choose from 500+ models at runtime, referencing live ELO rankings, intelligence scores, pricing, and latency — instead of relying on stale training-data hunches.

500+

Models behind one API (50+ free)

4.7M+

Voters powering Design Arena ELO

$10

Spend cap on OAuth key (7-day validity)

Local installs — fully remote server

The Selection Loop

How an agent routes a task to the right model at runtime.

Query benchmarks

Live ELO + intelligence scores

→

Check models-list

Price · context · latency

→

chat-send test

Validate response & cost

→

Route the task

Best model for the job

Tools On The Server

benchmarks

Artificial Analysis scores + Design Arena ELO rankings

models-list / endpoints

Pricing, context length, latency, throughput, data policy

chat-send

Send test prompts — billable, returns response and cost

rankings-daily

Trends based on daily token volume

Why developers like it

Real-time benchmarks & live pricing end the "same model everywhere" habit
Auto-route design tasks to the top-ELO model
Validate the choice with test inference first

The caveats

Positioned mainly as a development aid
Production apps should call the OpenRouter API directly
Very recent launch — long-term real-world evaluation still to come

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.