05-28-Daily AI News Daily

Daily Summary

OpenRouter's token volume surged 5x in six months and secured $113M Series B—the routing layer that "doesn't build models, just manages traffic" is officially capital's new darling.
Claude Code launched a security scanning plugin the same day with 127k stars in a single day. Anthropic isn't building a coding assistant anymore—it's going after VS Code's position.
Today's main thread is crystal clear: money and developers are flowing toward infrastructure, and Claude Code's ecosystem deserves close attention.

⚡ Quick Navigation

• 📰 Today’s AI News - Latest updates at a glance

💡 Tip: Want to experience the latest AI models mentioned here (Claude 4.5, GPT, Gemini 3 Pro) right now? No account? Grab one at Aivora —one minute setup, hassle-free support.

Today’s AI News

👀 One-Liner

OpenRouter’s token volume exploded 5x in six months and landed Series B; Claude Code’s plugin marketplace quietly launched security scanning—AI infrastructure money is consolidating in the “pipeline layer.”

🔑 3 Keywords

#PipelineLayerCashing #ClaudeEcosystemExpansion #DomesticModelPriceWar

🔥 Top 10 Headlines

1. OpenRouter Completes $113M Series B, Token Volume Surges 5x in Six Months

Six months ago, OpenRouter processed 50 trillion tokens per week. Now that number is 250 trillion. A clean 5x jump.

This “AI model aggregation router” just landed $113M Series B, and the logic is straightforward: more and more developers don’t want to be locked into a single model—they need one unified entry point to switch between GPT, Claude, and Gemini on the fly. OpenRouter is that “main pipeline.”

Some say the valuation is low—but flip the perspective: this is real infrastructure business. Don’t build models, just manage traffic, collect tolls while you sleep. The bottom layer of the token economy is just getting started.

2. Claude Code Launches Security Vulnerability Scanning Plugin, Auto-Scans While You Code

Used to be, finding security bugs meant either sharp-eyed code reviewers or alerts after production went live. Now Anthropic moved that checkpoint upstream.

Claude Code’s new security plugin automatically checks for vulnerabilities when you merge code and suggests fixes. All Claude Code users get it—search /plugins in the marketplace to install.

This move means more than “one more feature”—it signals Claude Code is serious about building a plugin ecosystem, upgrading itself from “AI coding assistant” to “full-stack development platform.” Cursor should be getting nervous.

3. Claude Code Official Repo Hits 127k Stars in a Single Day

Not a new product launch, but this number deserves its own mention.

Claude Code’s GitHub repo gained 127,043 stars today—basically the entire repo’s star count, meaning this project just exploded recently. A terminal-based agentic coding tool that understands codebases, handles Git workflows, and takes natural language instructions is becoming the hottest topic in dev circles.

Combined with the security plugin above, Anthropic’s investment in Claude Code doesn’t look like a “side project”—it looks like a serious play for the programming tools battlefield.

4. ElevenLabs Releases Music v2, Crystal-Clear Mandarin Pronunciation, Free Users Get 3 Songs

Suno has a chronic problem: singing Mandarin sometimes mangles the lyrics, sounds off. ElevenLabs’ Music v2 turned that pain point into a selling point—Mandarin pronunciation is crisp and flawless.

Overall musicality trails Suno slightly, but it’s solid first-tier quality. The real kicker: free users can use it and generate 3 songs, try it live at elevenlabs.io/app/music.

AI music competition is evolving from “can we generate it?” to “can we generate it accurately?” For creators making Mandarin content, this detail difference directly determines whether you can actually use it.

5. Tongyi Qianwen Upgrades “Photo Health Check,” Validated by Top-Tier Hospital Doctors

Snap a lab report, circle the abnormal values, AI gives reasoning and health advice—Tongyi Qianwen baked this workflow into the app.

New feature supports visual annotation and linked reasoning, plus it shows the reasoning process transparently instead of just dropping conclusions. Top-tier hospital doctors tested it and gave thumbs up.

The hardest part of health AI isn’t tech—it’s trust. Laying out the reasoning process for users to see is smarter design than just stacking parameters—users don’t need to believe AI is powerful, they just need to understand what it’s saying.

6. Bonsai Image 4B: Run AI Image Generation on 1.5GB RAM, Works on Phones

Running AI image generation on phones used to be fantasy. Bonsai Image 4B made it real.

Built on FLUX.2 Klein 4B, the 1-bit version runs with just 1.5GB active memory, quality close to the original FLUX. Comes in 1-bit (-1,+1) and Ternary (-1,0,+1) versions, models uploaded to HuggingFace.

On-device AI image generation matters because: no internet needed, no API fees, no server queue waits. Developers wanting to run creative tools locally should keep this on the radar.

7. Let Codex Scan Chat History, Extract Dev Preferences, Write Them Into Skills—Gets Smarter the More You Use It

Every time you finish with an AI coding tool, the experience evaporates. Next time you’re explaining “I like this code style” all over again.

Someone figured out a workaround: let Codex periodically scan this week’s chat history, auto-extract your dev habits and aesthetic preferences, write them into a Skill file. Now Codex gets smarter about you without needing retraining each session.

The core idea is “make tacit knowledge explicit”—your dev habits, code preferences, lessons learned, all crystallize into reusable context. The tool isn’t just helping you work; it’s learning who you are.

8. Designer Hand-Rolled Science Comic Website: Type a Species Name, Auto-Generate Beautiful Illustrated Guide

An internet designer, AI newbie, built their first website after picking up AI.

Simple feature: type a species name, auto-generate a beautiful science comic. URL is naturalart.zeabur.app, try it now. From screenshots, layout and illustration quality are seriously polished—doesn’t look like a first project.

The real value here isn’t the tool itself—it’s the trend it represents: AI is compressing “idea to product” from months to minutes. Designers don’t need to code, programmers don’t need to design, ideas themselves are becoming the premium asset.

9. AI Customer Service Gets 3D Virtual Body: Talks, Moves, Handles Interruptions, Full Code Included

The most frustrating thing about AI customer service isn’t wrong answers—it’s that “reading from a script” vibe: high latency, frozen expressions, completely ignores you if you interrupt mid-sentence.

A developer hooked a large model to a 3D virtual avatar via Morphic Cloud, nailing real-time interruption, natural expressions, and smooth conversation. Article includes full runnable Android code—apply for AppID and Secret, then run it.

The gap between “works” and “feels right” is exactly this layer of “humanity.” The commercial value is obvious: customer service, education, companionship apps—anywhere users need long conversations, 3D avatars become the new standard.

10. guizang-social-card-skill: AI Auto-Generates Xiaohongshu Posts, Travel Routes Marked Directly on Maps

The most tedious step in travel content: manually marking points on maps, screenshotting, pasting into design tools, tweaking endlessly. guizang-social-card-skill kills that workflow.

Feed it destination and route, AI auto-marks on the base map and embeds images—optimized for Xiaohongshu categories. Not just travel; every content category got specialized optimization. Skill launching soon.

For content creators, tools like this don’t shine because “AI writes your copy”—they shine because they automate away repetitive grunt work. The time you save is time you actually spend creating.

📌 Worth Watching

[Open Source] claude-code-harness: Let Claude Code Run Plan→Work→Review Loops Autonomously - 1821 Stars, shell-script-driven autonomous dev framework. Claude Code doesn’t just take orders—it plans, executes, and reviews itself. Worth trying if you want to free up your hands.

[Business] Xiaomi MiMo Follows DeepSeek, Permanent API Price Cut Up to 99% - Token usage per dollar up 5-8x at same price tier. Large model price wars shifted from “promotions” to “permanent repricing.” Projects you shelved over API costs are worth recalculating now.

[Product] RepoPrompt Author Hired by OpenAI, Tool Going Open Source Soon - Paid users getting Codex Credits as compensation. This Mac tool that packages entire code repos into XML for long-context models is worth tracking once it opens.

GPT 5.5 Pro Landed a Screenwriting Gig, Already Researching the Market Before Shooting Starts

Codex for code, Claude for copy, now someone’s outsourcing short-form drama scripts to AI too.

Someone had GPT 5.5 Pro first scout short-drama discussion forums to figure out what audiences actually want, then based on that research, wrote a short-drama script-generation Skill. Plan is to test results, then generate a few clips to see how they land.

Used to be: market research → strategy meeting → outline → script. Now that whole pipeline compresses into one Prompt. AI as screenwriter isn’t novel, but AI that “understands the audience” before writing? That’s more thoughtful than some human writers.

🔮 AI Trend Predictions

Claude Code Plugin Ecosystem Expands Rapidly, Becomes New Programming Tool Standard

• Prediction Timeline: July 2026
• Confidence: 75%
• Reasoning: Today’s news Claude Code Launches Security Scanning Plugin + Claude Code repo hit 127k stars in one day. Anthropic is systematically building a plugin marketplace. The security plugin is just the opening move—code review, test generation, deployment automation plugins will roll out densely, forming a VS Code-like moat.

AI Model API Prices Enter “Bargain Basement” Era, Mid-Tier Developers Win

• Prediction Timeline: Q3 2026
• Confidence: 80%
• Reasoning: Today’s news Xiaomi MiMo-V2.5 Permanent 99% Price Cut + DeepSeek pioneered permanent discounts, Xiaomi followed suit. Price wars shifted from “promotions” to “repricing.” Tier-one vendors will follow within months—API call costs will keep dropping through the rest of the year.

On-Device AI Image Generation Explodes on Mobile, Cloud API Dependency Drops

• Prediction Timeline: Q3 2026
• Confidence: 60%
• Reasoning: Today’s news Bonsai Image 4B Runs FLUX-Level Generation on 1.5GB RAM + Mobile chip compute keeps climbing, 1-bit quantization is mature, on-device inference is now within flagship phone range. Mobile local AI creation apps will flood the market in H2.

OpenRouter and Similar AI Routing Platforms See Rapid Valuation Growth, Become Next Funding Hotspot

• Prediction Timeline: Q3 2026
• Confidence: 65%
• Reasoning: Today’s news OpenRouter Lands $113M Series B + 5x token growth in six months validates “model routing layer” business model. As enterprise AI buying shifts from single-model to multi-model blends, aggregation API platforms become the next funding wave.

❓ Related Questions

How Do I Experience Claude Code?

Claude Code currently requires an Anthropic account and Claude Pro or Max subscription for full access. Domestic users face registration and payment barriers. The official repo is open source (github.com/anthropics/claude-code), but core features depend on Claude API calls.

Solution: Visit Aivora to grab a ready-to-use Claude account—instant delivery, hassle-free support, jump straight into Claude Code and the latest plugins.