OpenRouter · Launch Video Breakdown: Hook, Pacing & Motion Design
Live model intelligence inside your agent.
Scene-by-scene timeline & spoken transcript
The Hook
Introducing OpenRouter MCP
“Hey everyone, I'm here to talk about coding agents, open router models, and the Open Router MCP.”
- On screen
- (No on-screen text)
- Camera
- Medium close-up shot of the presenter, centered, with a shallow depth of field. The camera is static.
- Motion
- Direct address to camera, establishing personal connection. Minimal motion, focusing on the speaker's presence.
Product Reveal
Live Demo: Finding the Best Coding Models
“So let's say I'm building a coding agent and I want to find the best coding models right now under $2 million input. I can just ask Claude Code. And it's going to go out and call Open Router and find me the best coding models. So here we have Gemini 3.5 Flash, GLM 5.2, Gemini 3.1 Pro Preview, Qwen 3.7 Max, Kimi 2.7 Code, MiMo V2.5 Pro, and DeepSeek V4 Pro. And it's going to give me a breakdown of the coding score, the input cost, and the output cost. And it's going to give me some picks. So the best coding score in budget is Gemini 3.5 Flash or GLM 5.2. The best value is DeepSeek V4 Pro, and the cheapest decent is MiMo V2.5 Pro. And it's going to give me some notes about the overall coding leaders. And it's going to tell me where it got its sources from. So Artificial Analysis via Open Router. Now let's say I want to find the fastest provider for GLM 5.2. I can just ask Claude Code. And it's going to go out and call Open Router and find me the fastest provider. So here we have Friendli by far the fastest on both metrics. Throughput 152 tokens per second versus 40 to 50 for the rest. Latency 273 milliseconds versus 700 to 900 milliseconds. Uptime 99.9%, priced at the standard $1.40/$4.40. And it's going to give me some runners-up. So AkashML and Together. Now let's say I need a landing page for this SaaS. What are the best affordable models for designing the landing page?”
- On screen
- Search sessions, agents, files... ~/Code/openrouter-mcp-demo (0.024s) clear ~/Code/openrouter-mcp-demo (1m 48.00s) claude Claude Code v2.1.190 Welcome back! Tips for getting started Ask Claude to... What's new Bug fixes and... Added org-con... /release-note... Opus 4.8 (1M context) with me... API Usage Billing ~/Code/openrouter-mcp-demo FE:3000 bypass permissions on (shift+tab to cycle) for agents What are the top coding models right now, under $2/M input? Thought for 12s, called openrouter 2 times (ctrl+o to expand) Top coding models priced <= $2/M input tokens, ranked by Artificial Analysis coding index: Model Coding Input $/M Output $/M Gemini 3.5 Flash (high) 70.1 $1.50 $9.00 GLM-5.2 (max) 68.8 $1.40 $4.40 Gemini 3.1 Pro Preview 68.8 $2.00 $12.00 Qwen 3.7 Max 66.0 $1.25 $3.75 Kimi 2.7 Code 60.8 $0.75 $3.50 MiMo-V2.5-Pro 60.2 $0.60 $3.00 DeepSeek V4 Pro 59.4 $0.87 $1.74 Picks: Best coding score in budget: Gemini 3.5 Flash (70.1) or GLM-5.2 (68.8, cheaper output). Best value: DeepSeek V4 Pro - 59.4 coding at the lowest combined cost ($0.87 in / $1.74 out), 1M context. Cheapest decent: MiMo-V2.5-Pro ($0.60 in). Note: the overall coding leaders (Claude Fable 5 at 76.5, GPT-5.5 at 74.9, Opus 4.8 at 74.3) all exceed $2/M input. Sources: Artificial Analysis (artificialanalysis.ai) via OpenRouter. Cogitated for 17s What's the fastest provider for GLM-5.2? Called openrouter (ctrl+o to expand) Friendli - by far the fastest on both metrics: Throughput: 152 tok/s (p50), vs ~40-50 for the rest Latency (TTFT): 273 ms (p50), next best ~700-900 ms Uptime: 99.9%, priced at the standard $1.40/$4.40 Runners-up: AkashML (860 ms TTFT, 53 tok/s, and Together (48 tok/s, but slower TTFT and spikier). Worked for 7s I need a landing page for this SaaS. What are the best affordable models for designing the landing page?
- Camera
- The presenter is visible in a small circular inset at the top right, while the main screen displays a terminal interface. The terminal content scrolls and updates dynamically.
- Motion
- Picture-in-picture (PIP) for presenter overlaying a live terminal demo. Text appears to be typed and results populate dynamically, simulating real-time interaction.
Feature Teaser
Designing Landing Pages with OpenRouter
“For visual design specifically, Design Arena ELO is the right signal, not raw coding benchmarks. Filtering to affordable models, under $2 million input. So here we have GLM 5.2, GLM 5.1, Kimi 2.6, MiMo V2.5 Pro. And it's going to give me the ELO score, the input/output cost, and the speed. So GLM 5.2 is rank one, but it's slow. GLM 5.1 is also slow. Kimi 2.6 is very slow. MiMo V2.5 Pro is slow. So it's going to give me some picks. Best speed/quality balance, Gemini 3 Pro Preview, highest success rates, 64% website, 68% UI, 3x faster, just at the $2 input ceiling. Cheapest credible option, MiMo V2.5 Pro or even MiMo V2.5. For a landing page, I'd iterate quickly on Gemini 3 Pro Preview, fast feedback loops, strong design, then optionally do a final polish pass with GLM 5.2. Want me to draft the landing page itself? If so, tell me the product positioning and I can build it. Let's build the landing page with GLM 5.2 and show me the results. Booping. Tip: say fan out sub agents and Claude sends a team. Each one digs deep so nothing gets missed. Empty directory. I have no info on what this SaaS does. I need the basics before generating anything. User answered Claude's questions. What does this SaaS do? Product name plus one line pitch. Use a placeholder. What visual direction do you want? Clean and minimal. Generating with GLM 5.2 via Friendli, the fastest provider. Placeholder product: Helm, an AI ops console that turns infra logs into plain English answers. Calling Open Router. Booping. 47 seconds, 1.4k tokens. Tip: say fan out sub agents and Claude sends a team. Each one digs deep so nothing gets missed. Open Router ships an official Python SDK. Open Router Python SDK. I like GLM 5.2's design the best. Let's search docs tools for the Python SDK for Open Router so I can add it to my SaaS.”
- On screen
- Website arena: Model ELO Input/Output $/M Speed GLM-5.2 1357 (rank 1) $1.40 / $4.40 slow (~280s) GLM 5.1 1320 $0.98 / $3.08 slow Kimi 2.6 1318 $0.95 / $4.00 very slow (~460s) MiMo-V2.5-Pro 1313 $0.60 / $3.00 slow Best speed/quality balance: Gemini 3 Pro Preview - highest success rates (64% website, 68% UI), ~3x faster, just at the $2 input ceiling. Cheapest credible option: MiMo-V2.5-Pro ($0.60 in) or even MiMo-V2.5 ($0.14 in, 1304 ELO) for rapid iteration. For a landing page, I'd iterate quickly on Gemini 3 Pro Preview (fast feedback loops, strong design), then optionally do a final polish pass with GLM-5.2. Want me to draft the landing page itself? If so, tell me the product positioning and I can build it. Churned for 17s Let's build the landing page with GLM-5.2, and show me the results. Booping... Tip: Say "fan out subagents" and Claude sends a team. Each one digs deep so nothing gets missed. Empty directory - I have no info on what this SaaS does. I need the basics before generating anything. User answered Claude's questions: What does this SaaS do? (product name + one-line pitch) - Use a placeholder What visual direction do you want? - Clean & minimal Thought for 10s Generating with GLM-5.2 via Friendli (fastest provider). Placeholder product: Helm - an AI ops console that turns infra logs into plain-English answers. Calling openrouter... Booping... (47s - 1.4k tokens) Tip: Say "fan out subagents" and Claude sends a team. Each one digs deep so nothing gets missed. OpenRouter ships an official Python SDK (OpenRouter, Python) Thought for 3s, listed 1 directory I like GLM-5.2's design the best. Let's search docs tools for the Python SDK for OpenRouter so I can add it to my SaaS.
- Camera
- The presenter remains in the circular PIP. The terminal continues to display dynamic text generation, scrolling, and new data tables appearing.
- Motion
- Continuous live demo with text appearing as if typed, and results populating in real-time. Subtle highlighting and scrolling guide the viewer's attention.
Call to Action
Integrating OpenRouter MCP
“So here we have the Python SDK. We can see how to stream content, how to send messages, how to use the async variant. And it's going to give us some examples of how to use the SDK. So for your landing page generation flow, you'd pin the fast provider the same way the MCP call does. Pass provider only Friendli, allow fallbacks false to chat.send. Want me to scaffold a small generate landing page module in this repo wrapping that? Yes, scaffold generate landing page.py. So here we have the Claude MCP add transport HTTP open router, and then Claude MCP login open router. And that's how you get started with Open Router MCP.”
- On screen
- Streaming: stream = client.chat.send( model="z-ai/glm-5.2", messages=[{"role": "user", "content": "Write a story"}], stream=true ) for event in stream: content = event.choices[0].delta.content if event.choices else None if content: print(content, end="", flush=True) Async variant: await client.chat.send_async(...) inside async with OpenRouter(...). For your landing-page generation flow, you'd pin the fast provider the same way the MCP call does - pass provider={"only": ["friendli"], "allow_fallbacks": False} to chat.send. Want me to scaffold a small generate_landing_page.py module in this repo wrapping that? * Cogitated for 21s yes, scaffold generate_landing_page.py ~/Code/openrouter-mcp-demo (0.021s) clear ~/Code/openrouter-mcp-demo claude mcp add --transport http openrouter https://mcp.openrouter.ai/mcp; claude mcp login openrouter
- Camera
- The presenter remains in the circular PIP. The terminal continues to display code snippets and command-line interactions, demonstrating integration steps.
- Motion
- Code snippets and command-line instructions are presented with a typing animation, emphasizing the ease of integration. The overall motion is clean and functional.
Motion design primitives
Synthesizable in Remotion · key components of the developer-cli-live-demo template.
TerminalTypewriterEffect
Text appears character by character within a terminal window, simulating live typing. This is often accompanied by a blinking cursor.
+ Code construction recipe
Utilize Remotion's `Sequence` and `interpolate` functions to reveal text based on frame progress. Each character's opacity or visibility is animated from 0 to 1 over a short duration, with a slight delay for subsequent characters. A blinking cursor can be achieved with a `spring` animation on its opacity.
DynamicTablePopulation
Data tables populate row by row or cell by cell, with values appearing sequentially. This creates a sense of data being processed and presented in real-time.
+ Code construction recipe
Employ `Sequence` and `spring` or `interpolate` for each table row or cell. Animate the `y` position and `opacity` of each row/cell to slide in from the top or bottom, with staggered delays for each element to create a cascading effect. Use `delay` props on `Sequence` components for precise timing.
PresenterPIPOverlay
A circular video feed of the presenter is overlaid on the main content, typically in a corner, maintaining a personal connection while showcasing the product.
+ Code construction recipe
Render the presenter's video feed within a `Video` component. Apply `border-radius: 50%` and `overflow: hidden` for the circular mask. Position it using `absolute` positioning and `top`, `right` CSS properties. Ensure `z-index` is high enough for it to appear above the main content.
Style DNA & aesthetic system
Visual aesthetic
Typography
Motion
Special effects
Pacing & rhythm
Conceptual story rhythm
- 1
The Personal Introduction
Purpose. Establishes credibility and a direct connection with the audience through the founder's personal address.
Execution. A medium close-up of the founder speaking directly to the camera, creating an immediate sense of trust and authority.
- 2
Problem & Solution Demo
Purpose. Demonstrates the core problem (finding optimal models) and immediately showcases OpenRouter's efficiency as the solution.
Execution. Transition to a live terminal demo with the founder in a PIP, showing real-time queries and data output, highlighting speed and accuracy.
- 3
Advanced Use Cases & Features
Purpose. Expands on the product's capabilities beyond the initial problem, showcasing versatility and depth (e.g., landing page generation).
Execution. Continued terminal demo, introducing more complex queries and multi-step processes, with dynamic table and text generation.
- 4
Integration & Call to Action
Purpose. Guides users on how to integrate the product and encourages them to take the next step.
Execution. Displaying code snippets and command-line instructions for SDK integration, concluding with clear setup commands.
Remix in Animatiq Studio
Master remix prompt · agent-ready
Sample remix
Objective
Launch video for 'DataFlow CLI', a new command-line interface for real-time data pipeline management.How the recipe adapts
The video would open with the founder introducing DataFlow CLI. The main segment would feature a live terminal demo showing commands being typed to set up, monitor, and debug data pipelines. Dynamic tables would populate with real-time data throughput and error logs. The presenter's circular PIP would remain in the corner, guiding the narrative. The video would conclude with commands for installation and initial setup, maintaining the continuous, no-cut flow and the dark, technical aesthetic.Remix blueprint · developer-cli-live-demo
Scene durations (4 scenes, 150s)
Key Remotion components
Related Developer Tools Product Launches
Explore all Developer Tools launches →
xAI opens voice cloning on its API: build a custom voice in about two minutes or pick from 80+ voices in 28 languages.
Claude can now operate your Mac, opening apps, browsing and filling spreadsheets, as a research preview in Cowork and Claude Code.
Agentic CLI for coding, building apps, and automating workflows.