Gemini's 90% Coding Score Drops To 19%
Gemini 3.8 Flash costs $1.24 per task, not $0.75, input tokens are 1% of the bill. Cache writes, not reasoning, drain your wallet. Gemini 3.8 Flash launched September 2026 at an introductory $0.75 per million input tokens, but that sticker price masks the real cost of running a coding agent. Independent testing by Artificial Analysis shows a single task costs $1.24 at high effort, $0.93 at medium, and input tokens account for just one cent of that bill. The rest? Cache writes consume 45% of your invoice, cache hits another 33%, while Google's hourly context-storage meter ($0.50 per million tokens per hour) runs whether your agent is thinking or idle. The thinking_level parameter offers only three settings (low, medium, high); no minimal option exists, and medium is the only economically sensible choice, high effort costs 33% more for a single index point. Gemini 3.8 Flash scores 40 on Artificial Analysis's Intelligence Index at medium effort, identical to the older 3.7 Flash, yet runs 54 tokens per second slower (246 vs 300 t/s). The headline win, 90.8% on Terminal-Bench 2.1, collapses to 19.1% on the harder Terminal-Bench 4.0 suite, versus Claude Opus 5's 51.8%. OpenAI's GPT-5.6 Luna finishes the same tasks for just $0.18 per task, making Gemini 3.8 Flash seven times more expensive for negligible gains. On January 1, 2027, Google doubles all prices; the $1.24 bill becomes roughly $2.50. Google's own migration guide advises staying on 3.7 Flash for compute efficiency. This video is for builders choosing models for agentic workflows, teams budgeting AI infrastructure, and anyone comparing cost-per-task across Claude, GPT, and Gemini. Chapters: 0:00 The $0.75 headline nobody actually pays 0:36 Same price, three generations in six weeks 1:46 Real agents cost $1.24 per task 3:24 Where 99 cents of your bill vanishes 4:54 Cache writes are eating your lunch 6:19 Renting tokens by the hour 7:25 Three effort levels, zero budget control 8:33 Low effort is a losing bet 9:48 One index point, 33% markup 11:17 The one benchmark where it dominates 13:22 Google's own docs say stay behind 14:38 Same score, slower, cheaper rival wins 16:44 The benchmark Google actually advertises 17:48 Where it collapses against Opus 5 19:54 January first: prices double overnight 21:11 Luna does it for 18 cents 23:22 When Flash actually makes economic sense Tools & resources mentioned: - Artificial Analysis: https://artificialanalysis.ai - Google Cloud Gemini 3.8 Flash docs: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/guides/gemini-3-8-flash - Requesty pricing: https://www.requesty.ai/models/vertex/gemini-3.8-flash - Apidog Gemini pricing guide: https://apidog.com/blog/gemini-3-8-flash-pricing About The Stack The Stack helps you build with AI. Each video takes one tool, model, or workflow and shows how it works in a few focused minutes, with the real benchmarks and real costs. We go deep on Claude Code and Cursor for AI coding, AI agents and MCP servers, the open-source AI tools and GitHub repos most people miss, RAG and vector search, fine-tuning, and running local LLMs on your own machine with Ollama and LM Studio. We compare models like ChatGPT and Claude, test AI automation with Zapier, Make, and n8n, and flag the tools that actually ship. Subscribe for new breakdowns: https://www.youtube.com/@the-stack-ai?sub_confirmation=1 #gemini #ai agents #claude vs chatgpt
