MODEL-ROUTING

Cloudflare is merging Workers AI and AI Gateway into one control plane: unified /ai/ REST API, auto-created default gateways, AI Gateway credits spendable on Workers AI, and model-first routing that picks the provider for you. Here is what changes and what stays.

Steve Yegge's Flat Curve Society thesis turns the AI adoption question into an operating problem: teach people to use agents, then teach them to waste fewer tokens.

A practical guide to routing between Claude Opus 5, Sonnet 5, Haiku 4.5, GPT-5.6 Sol/Terra/Luna, and Kimi K3 based on task complexity, cost budget, and latency requirements - with decision frameworks and code examples.

Vercel AI Gateway gives you one API key and string model ids like moonshotai/kimi-k2.5 for hundreds of models. Here is how it works with the AI SDK, what BYOK and OIDC change, the honest tradeoffs, and who should actually use it.

A viral Hacker News thread about AI affordability points at the right problem, but developer teams need a more useful cost model: retries, cache misses, review time, routing, and failed loops.

Envoy AI Gateway 1.0 is production-ready. The useful question for builders is when an Envoy-based LLM gateway beats direct SDK calls, LiteLLM, OpenRouter, or a hosted AI gateway.

Sakana says Fugu Ultra stands with Fable, Mythos, GPT-5.5, Gemini, and Opus by orchestrating models instead of being one giant model. Here is what the benchmarks show, what is novel, and what still needs proof.

Sakana Fugu makes a timely argument for model routing: frontier performance should come from swappable systems, not a hard dependency on one proprietary API.

Sakana Fugu Ultra is not just another giant model. It is a learned orchestration layer that routes work across expert models, matches frontier benchmark claims, and makes a serious case for multi-model AI systems.

No single model wins every task anymore, and the companies that never trained one - Factory, Devin, Perplexity, Cursor, OpenCode - are turning that into a moat. This is how model routing works, why open weights and neoclouds make it cheap, and the honest counter-argument.

DeepSeek V4 Pro lands an 80.6 on SWE-bench Verified in Max reasoning mode at $0.66/$1.98 per million tokens off-peak, and Flash runs agent inner loops at $0.22/$0.66. Here is the worked cost math, the Flash-vs-Pro split, and a clear guide on when to route to DeepSeek instead of a frontier model.

Factory.ai shipped a router that auto-picks the model for each Droid session and fails over across providers. The vendor claims 20-25% lower token spend and 99.9%+ request reliability. Here is what the product actually does, which claims are vendor claims, and whether a router beats DIY routing for your team.

Perplexity launched a $200-a-month agent that coordinates 19 models and calls orchestration, not the model, the product. Here is the strategic case for why the durable, defensible layer in AI sits next to the labs, not inside them - and what 'token value per watt per user' actually means for builders.

OpenRouter Fusion turns multi-model panels into an API feature. The useful lesson is not to run every prompt through more models. It is to define when a task deserves an expensive second opinion.

Factory AI's Droid agent surfaces a new competitive front in coding tools: cost-per-completed-task. Here's what their architecture reveals about where the whole industry is heading.

A first-hand visit to DeepSeek HQ reveals something more interesting than benchmark scores: a 300-person company that treats AI as infrastructure, not eschatology - and what that means for API pricing everywhere.

OpenRouter gives you one API key for 300+ models, automatic fallbacks, and intelligent provider routing. Here is what it actually costs, how to set it up in five minutes, and when you should skip it entirely.

Microsoft's new in-house coding model matters less as a benchmark headline and more as a signal that Copilot is becoming a routing layer for cost, latency, ownership, and review quality.