npmrundeploy.com

Weekly AI Ranking

The 12 most notable AI products/models this week, ranked by real-world impact.

Week of 20/07 – 26/07/2026 · Hand-curated from discussion volume, news and hands-on impressions See methodology →

Audio — Listen to the ranking Read aloud automatically by your browser (~1 min)
12 tracked 2 new 3 up 0 down
New entry DeepSeek V4 New entry Kimi K3 Up Claude Up Cursor
Weekly AI Ranking — Week of 20/07 – 26/07/2026
Product Category
Rank 1 ChatGPT OpenAI The most widely used general-purpose AI assistant, leading in daily active users across web and mobile. Assistant Steady
Rank 2 Claude Anthropic A favorite among developers thanks to Claude Code and strong agentic ability to read and edit large codebases. Assistant Up
Rank 3 Gemini Google Deeply integrated into Google's ecosystem (Search, Workspace, Android) — a major distribution advantage. Assistant Steady
Rank 4 DeepSeek V4 DeepSeek Just shipped its stable release this week — a strong open reasoning model that keeps pressuring frontier-lab pricing. Coding New entry
Rank 5 Kimi K3 Moonshot AI Topped a major coding leaderboard this week; demand spiked so hard Moonshot had to pause new sign-ups. Coding New entry
Rank 6 GitHub Copilot GitHub / Microsoft Still the most widely deployed in-IDE coding assistant, riding its built-in distribution via GitHub/VS Code. Coding Steady
Rank 7 Cursor Anysphere A purpose-built AI-native IDE, growing fast among developers who want deeper agentic workflows than Copilot offers. Coding Up
Rank 8 Perplexity Perplexity AI An answer-first search engine with citations, remaining the most popular Google alternative. Research Steady
Rank 9 Midjourney Midjourney Inc. Remains the top pick among designers and artists for the aesthetic quality of generated images. Creative Steady
Rank 10 Runway Runway The most complete AI video creation toolkit for filmmakers and professional creators. Creative Steady
Rank 11 ElevenLabs ElevenLabs Leads in natural AI voice synthesis (TTS) and voice cloning, widely embedded in news-reader and audiobook apps. Creative Steady
Rank 12 NotebookLM Google A document research/summary assistant, well known for turning documents into an automatic conversational podcast. Assistant Up

This week's highlights

Ranking methodology

The npmrundeploy.com team hand-curates this table every week based on social/dev-community discussion volume, product launch news and hands-on trials — not an automated algorithm or benchmark. Ranks reflect the editorial team's judgment at publish time, and may differ from objective benchmark leaderboards like the Intelligence Index or OpenRouter below.

Editor: Đội ngũ npmrundeploy.com · Updated: 08/04/2026, 09:00 AM

Shareable image

9:16 ratio — optimized for Story / Reels / TikTok

🧠 Model intelligence index

Intelligence Index — Artificial Analysis

An independent ranking of AI model "intelligence," aggregated from a battery of benchmark tests run by Artificial Analysis.

1 Claude Opus 5 (max) Anthropic 61 Score
2 Claude Opus 5 (xhigh) Anthropic 60 Score
3 Claude Fable 5 (with fallback) Anthropic 60 Score
4 GPT-5.6 Sol (max) OpenAI 59 Score
5 Claude Opus 5 (high) Anthropic 59 Score
6 GPT-5.6 Sol (xhigh) OpenAI 58 Score
7 Kimi K3 Kimi 57 Score
8 Claude Opus 5 (medium) Anthropic 56 Score
9 GPT-5.6 Sol (high) OpenAI 56 Score
10 Claude Opus 4.8 (max) Anthropic 56 Score

Data source: Artificial Analysis — View source →

📊 Ranked by real-world usage

OpenRouter Rankings — this week's usage volume

Rankings based on real usage data (not benchmarks) from millions of requests flowing through OpenRouter each week, broken down across 7 task categories.

Week of 2026-07-25

🏆 Top models by overall usage

  1. 1 DeepSeek V4 Flash deepseek 7.22T tokens +13%
  2. 2 MiMo-V2.5 xiaomi 6.3T tokens +40%
  3. 3 Hy3 tencent 4.82T tokens +22%
  4. 4 DeepSeek V4 Pro deepseek 3.28T tokens +3%
  5. 5 GLM 5.2 z-ai 2.89T tokens +12%
  6. 6 Nemotron 3 Ultra (free) nvidia 2.43T tokens +4%
  7. 7 MiniMax M3 minimax 1.96T tokens +4%
  8. 8 GPT-5.6 Luna openai 1.95T tokens +465%
  9. 9 Step 3.7 Flash stepfun 1.66T tokens +14%
  10. 10 Kimi K3 moonshotai 1.42T tokens +16%

⚡ Quick start via OpenRouter

OpenRouter lets you call almost every model on the market through a single API endpoint (OpenAI-compatible), without registering separately with each provider.

curl https://openrouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "<model-id>", "messages": [{"role":"user","content":"Xin chào"}]}'

See full docs → · openrouter.ai/models →

Text (LLM)

Model Usage / week
  1. 1 MiMo-V2.5 9.95T tokens +9%
  2. 2 DeepSeek V4 Flash 5.81T tokens +9%
  3. 3 Hy3 (free) 4.93T tokens +53%
  4. 4 GLM 5.2 3.58T tokens +1%
  5. 5 DeepSeek V4 Pro 3.01T tokens +16%

Quick take: MiMo-V2.5 (Xiaomi) leads this week by token volume — a fast-rising open model; Tencent's Hy3 (free) jumped 53%, showing strong demand for free-tier trials.

Image

Model Usage / week
  1. 1 Nano Banana (Gemini 2.5 Flash Image) 1.3M requests +5%
  2. 2 Nano Banana 2 (Gemini 3.1 Flash Image) 791K requests +16%
  3. 3 Nano Banana 2 (Gemini 3.1 Flash Image Preview) 727K requests +11%
  4. 4 Nano Banana Pro (Gemini 3 Pro Image Preview) 520K requests +0%
  5. 5 GPT Image 2 515K requests +13%

Quick take: Google's "Nano Banana" family (Gemini 2.5/3.1 Flash Image) takes 4 of the top 5 image-gen spots — a clear distribution edge via Gemini, ahead of OpenAI's GPT Image 2 in 5th.

Embeddings

Model Usage / week
  1. 1 Text Embedding 3 Small 87.5M requests +10%
  2. 2 Qwen3 Embedding 8B 68.7M requests +3%
  3. 3 bge-m3 16.2M requests +55%
  4. 4 Text Embedding 3 Large 14.1M requests +21%
  5. 5 Embed V1 0.6B 12.3M requests +20%

Quick take: OpenAI's Text Embedding 3 Small remains #1 by a wide margin (87.5M requests/week); Qwen3 Embedding 8B holds a close 2nd, while Gemini Embedding variants are the fastest growers in the 6-8 range.

Rerank

Model Usage / week
  1. 1 Llama Nemotron Rerank VL 1B V2 (free) 1.49M requests +26%
  2. 2 Rerank v3.5 689K requests +6%
  3. 3 Rerank 4 Fast 582K requests +141%
  4. 4 Rerank 4 Pro 310K requests +10%

Quick take: NVIDIA's free Llama Nemotron Rerank leads by volume, while Cohere completely dominates the rest of the paid top 4 — no other provider breaks in.

Video

Model Usage / week
  1. 1 Veo 3.1 Fast 37K requests +17%
  2. 2 Seedance 2.0 25K requests +13%
  3. 3 Veo 3.1 Lite 15K requests +35%
  4. 4 Veo 3.1 13K requests +30%
  5. 5 Seedance 2.0 Fast 11K requests +28%

Quick take: Google's Veo 3.1 Fast leads video generation, with ByteDance's Seedance 2.0 close behind in 2nd — these two dominate share, and the faster "Fast/Lite" variants outpace the full versions.

Speech (TTS)

Model Usage / week
  1. 1 Kokoro 82M 604K requests +78%
  2. 2 Gemini 3.1 Flash TTS Preview 415K requests +9%
  3. 3 Grok Voice TTS 1.0 326K requests +95%
  4. 4 MAI-Voice-2 31K requests +35%
  5. 5 Aura-2 8K requests +175%

Quick take: Kokoro 82M — a compact open-source TTS model by hexgrad — leads usage; x-AI's Grok Voice TTS is growing fast (+95%), and Deepgram's Aura-2 surged +175% despite still ranking lower.

Transcription

Model Usage / week
  1. 1 Whisper Large V3 3.6M requests +238%
  2. 2 Whisper Large V3 Turbo 1.57M requests +57%
  3. 3 GPT-4o Mini Transcribe 675K requests +33%
  4. 4 Qwen3 ASR Flash 247K requests +7%
  5. 5 Whisper 1 174K requests +54%

Quick take: OpenAI's Whisper family sweeps the top 2 spots; Whisper Large V3 usage spiked +238% this week, showing surging API transcription demand even as challengers like Qwen3 ASR Flash keep pace.

Data source: OpenRouter — View source →

🎯 Ranked by aggregated benchmarks

Onyx LLM Leaderboard — weekly summary

Onyx aggregates many benchmarks (reasoning, coding, math, agentic, SWE-bench, chat) and sorts models into 5 tiers, S/A/B/C/D — S being the clear leaders.

Data as of

S Claude Fable 5 · GPT-5.6 Sol · Kimi K3
A Claude Opus 4.8 · GLM-5.2 · Claude Sonnet 5 · GPT-5.5 · DeepSeek-V4-Pro · Kimi K2.6 · MiniMax M3 · Hunyuan Hy3 · DeepSeek V3.2
B Muse Spark 1.1 · Gemini 3.5 Flash · Grok 4.5 · GPT-5.6 Terra · Qwen3.7-Max · ERNIE 5.1 · Nova 2 Pro · Step-3.7-Flash · Nemotron 3 Ultra · Gemini 3.1 Pro · Qwen 3.5 · DeepSeek R1 · Step-3.5-Flash · MiMo-V2-Flash
C MiMo-V2.5-Pro · GPT-5.6 Luna · GPT-oss 120B · Nemotron Ultra 253B
D Grok 3 · Llama 4 Maverick

Data source: Onyx LLM Leaderboard — View source → · Data as of