In this "Input/Output" episode of the Inside Java Podcast, Nicolai Parlog talks to Martin Chalupa and John Rose about Java's ahead-of-time capabilities and how Netflix's training run approach is the ideal one.

If you look at the actual share of input/output tokens though, it's ~6:1. Perhaps openrouter is assuming a far lower cache hit ratio than is true in practice. Either way, it's an interesting case of the router dynamics distorting the natural pricing for the inference providers! I apologize to companies currently exploiting this if I'm giving away valuable alpha :^) (5/5)

Gemini 4 Argon’s standard pricing is $4/$20 per 1M input/output tokens, with a 95% discount for cached input tokens. Compared to Gemini 3.1 Pro Preview, this is an increase in token cost, and a larger discount on cached input. At standard pricing, Gemini 4 Argon’s Cost per Task is $3.98, ~1.2x that of GPT-6 Astra ($3.26, max). At discounted pricing however, the Cost per Task is $1.99, making it a 40% reduction from GPT-6 Astra (max)

Second, I suspect openrouter is just overweighing input prices for routing purposes. Experimenting with GLM 5.3 between inferencenet (0.08$/5$) and Novita (0.42$/1.32$) , it routes to inferencenet 70% of the time, which corresponds to a 95/5 weighting of input/output. (4/5)

🚨 GPT-6 Sol is HALF the price of GPT-5.6 Sol. Standard API pricing per 1M tokens (input / output): Sol: $4 / $20 → $2 / $10 Luna: $0.20 / $1.20 → $0.10 / $0.50 That’s 50% cheaper input and nearly 60% cheaper output for Luna.

most people forget there are two vectors to optimize for to reduce model cost: 1) reducing the input/output costs, increasing cache hit rates, batching etc 2) packing more intelligence per token (fewer tokens for same task) inference cost = price per token × tokens per task optimizations around the second is underrated and something we’ll only see more of. compression and concision is intelligence.