Independent Research ยท July 11, 2026

Compare AI Models by Price and Quality July 2026

A quantitative comparison of 28 frontier and budget AI models across price (blended official provider API cost per 1M tokens) and quality (Artificial Analysis Intelligence Index v4.1). An independent analysis by YOU-DO-NOTHING - the execution layer for systematic growth. See how we build growth systems →

Quality: Artificial Analysis Intelligence Index v4.1 (July 2026) - aggregates GDPval-AA v2, Terminal-Bench v2.1, SciCode, GPQA Diamond, Humanity's Last Exam & 4 others. Higher is better (0-100 scale).
Pricing: Official provider API pages (Anthropic, OpenAI, Google, xAI, DeepSeek, Z.AI, MiniMax, Alibaba, Moonshot, Xiaomi) - verified July 11, 2026.
Blended cost = 75% input + 25% output per 1M tokens. Some AA Index scores are estimated from benchmark proxies where official scores are pending. · you-do-nothing.com

How we maintain this
Methodology

Pricing is pulled directly from each provider's official API page, not from aggregators or marketplaces. Quality is measured by the Artificial Analysis Intelligence Index v4.1, an independent benchmark aggregating nine evaluations across coding, reasoning, and agentic tasks. We verify pricing weekly, add new frontier models within 72 hours of public release, and remove deprecated models within two weeks of provider end-of-life. No affiliate links, no referral codes, no kickback. The comparison exists because we needed it ourselves.

Update history
All changes since launch
2026-07-11 Major rebuild. New quality metric, official pricing, model list pruned to 28 Major ›
Switched quality metric from LMSYS Chatbot Arena Elo to Artificial Analysis Intelligence Index v4.1 (0-100 scale) for broader model coverage and more objective scoring. Replaced OpenRouter blended pricing with official provider API pricing sourced directly from each lab's pricing page. Pruned model list from 51 to 28 current models. Removed 23 outdated entries (Claude Opus 4.5/4.6/4.7, GPT-5 Pro/5.2 Pro, Gemini 2.5, Llama 3.x, Mistral, Gemma, old Qwen3). Added 2026 frontier: GPT-5.6 Sol, Grok 4.5, Claude Sonnet 5, Gemini 3.5 Flash, Qwen3.7 Max/Plus, Kimi K2.7 Code, MiMo V2.5 Pro/V2.5. Added context window as tooltip field. Added this maintenance log.
2026-07-11 Pricing correction. MiniMax M3 and DeepSeek V4 Pro updated to current rates Pricing ›
Fixed MiniMax M3 from $0.60/$2.40 to $0.30/$1.20 after the July 10 price halving. Fixed DeepSeek V4 Pro from $1.74/$3.48 to $0.435/$0.87. The 75% launch discount is now the permanent standard list price per DeepSeek's May 22 announcement.
2026-07-11 Metric migration. Why LMSYS Elo was replaced with Artificial Analysis Metric ›
LMSYS Chatbot Arena Elo lacked comprehensive coverage of Chinese open-weight models (GLM, Kimi, MiniMax, MiMo, Qwen). Several Elo scores on the previous version were estimated proxies presented as real. Artificial Analysis Intelligence Index v4.1 covers all 28 models on this page with independent, reproducible evaluations across nine benchmarks including Terminal-Bench v2.1, SciCode, GPQA Diamond, and Humanity's Last Exam.
2026-07-10 Verification pass. All 28 model prices cross-checked against official provider pages Verified ›
Cross-checked all 28 model prices against official provider API pages: anthropic.com/pricing, developers.openai.com, ai.google.dev, x.ai/api, api-docs.deepseek.com, open.bigmodel.cn, minimax.io, help.aliyun.com, platform.moonshot.ai, xiaomi.com/mimo. All 28 verified within 24 hours.
2026-07-01 Initial launch. 51 models, OpenRouter pricing, LMSYS Elo, Chart.js scatter plot Launch ›
First version of the AI Model Comparison page. 51 models across S/A/B+/B/C/Free tiers. Pricing sourced from OpenRouter API (blended input/output). Quality measured by LMSYS Chatbot Arena Elo ratings (900-1550 scale). Built as a standalone static HTML page with Chart.js scatter plot, tier filtering, and four stat cards (Best Value, Highest Quality, Cheapest Paid, Best All-Rounder). Published at you-do-nothing.com/tools/ai-model-comparison.html.