Signal

Ideas shaping the agent economy.

Gemini 3.8 Flash listed at $0.75 per 1M input and $3.75 per 1M output, beside a corrected Gemini 3.6 Flash price

LatestSep 2026ModelsRouting

Gemini 3.8 Flash, and the Price We Had Wrong

Gemini 3.8 Flash is live on BlockRun at $0.75/$3.75 per 1M. Listing it made us re-read the price sheet, where Gemini 3.6 Flash had been billing twice what it should — and that stale number turned out to be hiding two fallbacks that were overcharging callers.

Read the article

Illustration of a routing valve splitting one stream of tokens between a large expensive pipe and several smaller efficient pipes

Aug 2026Routing

When LLM Routing Saves Money — and When It Doesn't

LLM routers are sold with savings numbers that are technically true and mostly misleading. The honest version: the three places routing savings actually come from, measured numbers with confidence intervals, and the cases where a router won't help you.

An abstract model routing system filtering an Agent request through constraint gates into a qualified model portfolio

Aug 5, 2026RoutingBenchmarks

A Constraint-First Model Router for AI Agents

How a local, inspectable Router improved verified Agent success while lowering normalized cost per successful task—and what the evidence still does not prove.

OpenClaw cost and reliability — four structural problems ClawRouter fixes via local routing

Jun 2026RoutingAgents

OpenClaw Is Burning Your Budget and Stalling Your Agents — 4 Fixes From the Issue Tracker

We read OpenClaw's issue tracker for the problems users hit in production: cost spikes from background work, provider outages that deadlock sessions, no spend ceiling for autonomous agents, one model for every task. ClawRouter's local routing, multi-provider fallback chains, and per-run cost caps fix all four.

ClawRouter decision layer — 15-dimension scoring, tier classification, quality-first fallbacks

Apr 2026Routing

Inside ClawRouter's Decision Layer: Real-Time Quality vs Cost

How ClawRouter actually decides which model handles your prompt — the 15-dimension classifier, sigmoid confidence calibration, quality-first fallback chains, and why 'cheapest sufficient' is the wrong framing.

ClawRouter routing diagram on Base — agent-native LLM router with 15-dimension classifier and USDC payment

Apr 2026Routing

ClawRouter: The Only LLM Router Built for Autonomous Agents (Open Source, USDC on Base)

Agents can't sign up for accounts. Agents can't enter credit cards. Agents can only sign transactions. ClawRouter is the only open-source LLM router that lets AI agents operate independently — 15-dimension classifier, sub-1ms local routing, 92% blended cost savings, USDC via x402 on Base & Solana. 9 models free, no signup.

ClawRouter vs OpenRouter — 100 OpenClaw issues analyzed, every structural problem solved

Mar 2026Routing

We Read 100 OpenClaw Issues About OpenRouter. Here's What We Built Instead.

OpenRouter is the most popular LLM aggregator — and the source of 100+ issues in OpenClaw's tracker. Broken failover, API key leakage, billing opacity, model ID mangling. ClawRouter solves all of them.

Smart LLM routing — 46 models, 14 dimensions, sub-1ms classification

Mar 2026RoutingModelsBenchmarks

LLM Router Benchmark: 46 Models, 8 Providers, Sub-1ms Routing

We benchmarked 46 LLM models from 8 providers through a production payment pipeline. Speed and intelligence are poorly correlated. Here's how we built a 14-dimension classifier that routes requests in under 1ms.

OpenClaw standalone vs OpenClaw + ClawRouter — full comparison across cost, context, error handling, and budgeting

Mar 2026Routing

AI Agent API Costs: How ClawRouter Cuts LLM Spending by 500×

AI agents on OpenClaw burn through API budgets with no cost awareness. ClawRouter's 7-layer token compression, automatic tier routing, and wallet-based budget caps cut LLM API costs 50–500×.