ModelsFugu Max: Sakana AI Bets on Orchestration for Cheap Frontier Work
Fugu Max is an orchestrator that sends each task to the leanest capable model in a pool that includes NVIDIA Nemotron. What it costs, what it claims, and its limits.
ReadNotes on models, agents, and what inference costs
Practical writing for developers building with large language models — how they work, how to pick one, and how to keep the bill predictable.
Models
ModelsFugu Ultra v2: What Sakana's Top Orchestrator Claims and Costs
Fugu Ultra v2 is Sakana AI's quality-first multi-agent model, released in September 2026. Its benchmark claims, its long-context pricing step and when it beats Max.
ReadModelsFugu Ultra: Sakana AI's First Quality-First Orchestrator
Fugu Ultra launched in June 2026 as a multi-agent system behind one API, got a v1.1 pool update in July, and was followed by v2 in September. What that history means.
Read
ModelsGemini 2.5 Flash Guide: The Model That Made Thinking Optional
Gemini 2.5 Flash let you switch reasoning on and off per request. What it costs, where it sits between Flash-Lite and Pro, and whether to build on it in late 2026.
Read
ModelsGemini 2.5 Flash Image (Nano Banana): Deprecated, Now What
Gemini 2.5 Flash Image, the original Nano Banana, is deprecated. What it did well, what it cost, and how to move to its successors without surprises.
ReadModelsGemini 2.5 Flash-Lite Guide: The Cheapest 2.5 Model, and Its Limits
Gemini 2.5 Flash-Lite costs $0.10 in and $0.40 out per million tokens with thinking off by default. The jobs it suits, the arithmetic, and when to move to 3.1.
Read
ModelsGemini 2.5 Pro Guide: Pricing Tiers, Retirement and Migration
Gemini 2.5 Pro made thinking the default for Google's flagship. What it still does well, the 200K pricing cliff, its retirement timeline, and how to move off it.
ReadModelsGemini 3 Flash Preview: Pro-Grade Reasoning at a Flash Price
Google's December 2025 Flash model brought a 1M window, four thinking levels and a 78% SWE-bench Verified claim. What it costs and how to plan around a preview.
ReadModelsGemini 3.1 Flash-Lite Guide: Thinking Levels at High Volume
Google's first Gemini 3 Flash-Lite costs more than 2.5 Flash-Lite but less than 2.5 Flash. What it claims, how its thinking levels work, and where it fits a router.
ReadModelsGemini 3.1 Pro Custom Tools: When Your Agent Overuses Bash
A separate Gemini 3.1 Pro Preview endpoint that prefers your registered tools over a generic shell. What it changes, what it costs, and how to tell if you need it.
ReadModelsGemini 3.1 Pro Preview: When the Pro Tier Still Earns Its Price
Gemini 3.1 Pro has been in preview since February 2026 while cheaper Flash models caught up on agent work. Where it still leads, what it costs, and how to plan around it.
Read
ModelsGemini 3.5 Flash-Lite: A Subagent Model Worth Its Price Rise
Gemini 3.5 Flash-Lite costs more than 3.1 Flash-Lite but nearly doubles its agentic scores. Where it fits as a subagent, and when to pay for 3.6 Flash instead.
ReadShowing 49–60 of 731 articles