OpenRouter Weekly: Hy3 preview (free) leads with 3.43T tokens (week of 2026-05-05)

Steady week on OpenRouter. Hy3 preview (free) retains #1 on the LLM Leaderboard, and the top three on each ranking remained intact. Smaller reshuffles below: see the per-area breakdowns for the movers.

LLM Leaderboard

The big picture: which models AI builders are paying to run this week.

1. Hy3 preview (free) by tencent: 3.43T tokens ↑525%
2. Kimi K2.6 by moonshotai: 1.79T tokens ↑5%
3. Claude Sonnet 4.6 by anthropic: 1.34T tokens ↑1%
4. Gemini 3 Flash Preview by google: 965B tokens ↑6%
5. Claude Opus 4.7 by anthropic: 926B tokens ↑21%

Market Share

Vendor share of total OpenRouter traffic: who's winning the platform race.

1. tencent 793B tokens ↑16.4%
2. google 681B tokens ↑14.1%
3. anthropic 566B tokens ↑11.7%
4. deepseek 488B tokens ↑10.1%
5. openai 457B tokens ↑9.4%

Benchmark Leaders (AA Index)

Independent benchmark scores via Artificial Analysis. High AA Index = stronger reasoning, but cost matters too.

1. GPT-5.5 (xhigh) by openai: 60.2 AA Index
2. Claude Opus 4.7 (Adaptive Reasoning, Max Effort) by anthropic: 57.3 AA Index
3. Gemini 3.1 Pro Preview by google: 57.2 AA Index
4. GPT-5.4 (xhigh) by openai: 56.8 AA Index
5. Kimi K2.6 by moonshotai: 53.9 AA Index

Fastest Models

Throughput champs: pick these for latency-sensitive apps where speed beats raw quality.

1. gpt-oss-120b
2. gpt-oss-safeguard-20b
3. gpt-oss-20b
4. Llama 3.1 8B Instruct
5. GLM 4.7

Top Coding Models

If you're shipping coding workflows on OpenRouter, this is what other builders chose this week.

1. Hy3 preview (free) by tencent: 509B tokens ↑30.7%
2. Kimi K2.6 by moonshotai: 372B tokens ↑22.4%
3. Step 3.5 Flash by stepfun: 78.6B tokens ↑4.7%
4. DeepSeek V4 Pro by deepseek: 66.6B tokens ↑4%
5. Claude Sonnet 4.6 by anthropic: 54.8B tokens ↑3.3%

Top Models for English

Most-used models for English content this week: the multilingual leaders.

1. Hy3 preview (free) by tencent: 695B tokens ↑14.4%
2. Kimi K2.6 by moonshotai: 466B tokens ↑9.6%
3. Claude Sonnet 4.6 by anthropic: 247B tokens ↑5.1%
4. MiniMax M2.7 by minimax: 173B tokens ↑3.6%
5. Gemini 3 Flash Preview by google: 171B tokens ↑3.5%

Top Models for Python

Which models are getting picked for Python work right now.

1. Hy3 preview (free) by tencent: 190B tokens ↑19.3%
2. Kimi K2.6 by moonshotai: 70.8B tokens ↑7.2%
3. Claude Sonnet 4.6 by anthropic: 39.7B tokens ↑4%
4. MiniMax M2.7 by minimax: 38.4B tokens ↑3.9%
5. DeepSeek V4 Flash by deepseek: 32.2B tokens ↑3.3%

Top Models for 1K - 10K tokens Prompts

For 1K - 10K tokens-sized prompts (the bulk of typical traffic), here's the lineup.

1. Gemini 2.5 Flash Lite by google: 27.9M requests ↑10.8%
2. Grok 4.1 Fast by x-ai: 22.8M requests ↑8.8%
3. Gemini 2.5 Flash by google: 18.5M requests ↑7.2%
4. Gemini 3 Flash Preview by google: 12.5M requests ↑4.8%
5. gpt-oss-120b by openai: 11.9M requests ↑4.6%

Top Models for Tool Calls

If your stack uses tool calls / function calling, these models are getting the most invocations.

1. Hy3 preview (free) by tencent: 10.4M tokens ↑16.4%
2. Kimi K2.6 by moonshotai: 4.08M tokens ↑6.4%
3. GPT-4o-mini by openai: 3.89M tokens ↑6.1%
4. Gemini 3 Flash Preview by google: 3.54M tokens ↑5.6%
5. Gemini 2.5 Flash by google: 3.11M tokens ↑4.9%

Top Image Models

Image-generation through OpenRouter: most-served models this week.

1. Gemini 2.5 Flash Lite by google: 40.4M images ↑36.9%
2. GPT-4.1 Mini by openai: 9.96M images ↑9.1%
3. Gemini 2.5 Flash by google: 8.46M images ↑7.7%
4. Gemini 3 Flash Preview by google: 5.66M images ↑5.2%
5. Gemini 3.1 Flash Lite Preview by google: 3.59M images ↑3.3%

Top Audio-Input Models

Audio-input (transcription, voice-in) leaders.

1. Gemini 2.5 Flash by google: 151K prompts ↑33.7%
2. Gemini 3 Flash Preview by google: 88K prompts ↑19.8%
3. Gemini 2.0 Flash Lite by google: 29K prompts ↑6.5%
4. Gemini 3.1 Flash Lite Preview by google: 25K prompts ↑5.6%
5. Gemini 2.5 Flash Lite by google: 21K prompts ↑4.7%

Top Apps on OpenRouter

Useful as social proof when picking your stack: the largest public apps and agents that opt into OpenRouter usage tracking.

Most Popular

1. **opting into](https://openrouter.ai/docs/app-attribution) usage tracking on OpenRouter. View more →

![Favicon for https://openclaw.ai/: 10.8T tokens: OpenClaw OpenClaw is an open-source AI agent that connects to your messaging apps and takes real actions on your behalf, from running commands and browsing the web to managing files and sending emails. 2. Hermes Agent: 5.83T tokens: Hermes Agent is an open-source, self-improving AI agent by Nous Research that runs persistently with memory across sessions, and builds reusable skills from experience. It comes with 40+ built-in tools, including web search, browser automation, and vision, plus scheduled automations and subagents. 3. Kilo Code: 5.54T tokens: Kilo Code is an open-source AI coding agent that works across VS Code, JetBrains, and CLI to help developers ship code faster with agentic workflows. 4. Claude Code**: 3.62T tokens: Claude Code is Anthropic's agentic coding tool that reads your entire codebase, plans and executes changes across files, runs tests, and iterates on failures, all from natural language prompts.

Trending

Fastest-growing apps on OpenRouter this week: early signals of which builder workflows are breaking out.

1. pi 224B tokens ↑719%
2. Incepedia synthetic pretraining data 22.4B tokens ↑6,885%
3. pfp-analyze 17.3B tokens ↑669,609%
4. extra.email 23B tokens ↑223%
5. VidMuse 56.6B tokens ↑37%
6. HotWordDeal 15B tokens ↑14,036%
7. Mira 44.7B tokens ↑33%
8. DealerCRM Image Verifier 9.26B tokens ↑23,726%