Daily snapshot · immutable permalink
AI model usage rankings - August 4, 2026
As published on August 4, 2026, DeepSeek: DeepSeek V4 Flash 0423 led real OpenRouter usage share at 20.7%. Qwen: Qwen3.8 Max, Qwen: Qwen3.7 Flash, Poolside: Laguna XS 2.1 (free), OpenAI: GPT-5.4 Mini (batch) entered the leaderboard for the first time. Anthropic: Claude Opus 4.7 (batch) moved 8 places up since August 3, 2026.
- #1 DeepSeek: DeepSeek V4 Flash 0423 - 20.7% usage share
- #2 Xiaomi: MiMo-V2.5 - 9.8% usage share
- #3 Tencent: Hy3 - 9.6% usage share
- #4 OpenAI: GPT-5.6 Luna (batch) - 6.0% usage share
- #5 Z.ai: GLM 5.2 (batch) - 5.7% usage share
- #6 DeepSeek: DeepSeek V4 Pro - 4.3% usage share
- #7 NVIDIA: Nemotron 3 Ultra (free) - 3.3% usage share
- #8 Poolside: Laguna S 2.1 (free) - 3.0% usage share
- #9 MiniMax: MiniMax M3 (batch) - 2.5% usage share
- #10 MoonshotAI: Kimi K3 - 2.1% usage share
| # | Model | Usage share | Rank change |
|---|---|---|---|
| 01 | DeepSeek: DeepSeek V4 Flash 0423 | 20.7% | no change |
| 02 | Xiaomi: MiMo-V2.5 | 9.8% | +1 (+1.7pts) |
| 03 | Tencent: Hy3 | 9.6% | -1 (0pts) |
| 04 | OpenAI: GPT-5.6 Luna (batch) | 6.0% | no change |
| 05 | Z.ai: GLM 5.2 (batch) | 5.7% | +1 (+1.2pts) |
| 06 | DeepSeek: DeepSeek V4 Pro | 4.3% | -1 (-0.3pts) |
| 07 | NVIDIA: Nemotron 3 Ultra (free) | 3.3% | no change |
| 08 | Poolside: Laguna S 2.1 (free) | 3.0% | no change |
| 09 | MiniMax: MiniMax M3 (batch) | 2.5% | no change |
| 10 | MoonshotAI: Kimi K3 | 2.1% | +4 (+0.4pts) |
| 11 | Claude Opus 5 (batch) | 1.9% | no change |
| 12 | StepFun: Step 3.7 Flash | 1.9% | -2 (-0.3pts) |
| 13 | Ling-3.0-flash | 1.8% | -1 (-0.1pts) |
| 14 | Anthropic: Claude Sonnet 5 (batch) | 1.8% | -1 (-0.1pts) |
| 15 | Google: Gemini 3 Flash Preview (batch) | 1.4% | no change |
| 16 | Anthropic: Claude Sonnet 4.6 (batch) | 1.4% | no change |
| 17 | OpenAI: GPT-5.6 Terra (batch) | 1.2% | no change |
| 18 | Google: Gemini 2.5 Flash Lite (batch) | 0.9% | no change |
| 19 | Google: Gemini 2.5 Flash (batch) | 0.9% | +3 (0pts) |
| 20 | Anthropic: Claude Opus 4.8 (batch) | 0.8% | -1 (-0.1pts) |
| 21 | OpenAI: GPT-5.6 Sol (batch) | 0.8% | -1 (-0.1pts) |
| 22 | OpenAI: gpt-oss-120b | 0.8% | +2 (+0.1pts) |
| 23 | Google: Gemini 3.1 Flash Lite (batch) | 0.8% | -2 (-0.1pts) |
| 24 | Xiaomi: MiMo-V2.5-Pro | 0.8% | -1 (-0.1pts) |
| 25 | OpenAI: GPT-5.6 Luna Pro (batch) | 0.7% | +1 (+0.1pts) |
| 26 | DeepSeek: DeepSeek V3.2 | 0.6% | +2 (+0.1pts) |
| 27 | Google: Gemma 4 31B (free) | 0.6% | no change |
| 28 | SpaceXAI: Grok 4.5 | 0.5% | +1 (0pts) |
| 29 | Google: Gemini 3.6 Flash (batch) | 0.5% | -4 (-0.2pts) |
| 30 | NVIDIA: Nemotron 3 Super (free) | 0.4% | no change |
| 31 | Google: Gemma 4 26B A4B (free) | 0.4% | no change |
| 32 | Anthropic: Claude Fable 5 (batch) | 0.4% | +1 (0pts) |
| 33 | Anthropic: Claude Haiku 4.5 (batch) | 0.4% | -1 (0pts) |
| 34 | Qwen: Qwen3.8 Max New | 0.4% | - |
| 35 | Google: Gemini 3.5 Flash (batch) | 0.3% | +2 (0pts) |
| 36 | inclusionAI: Ling-2.6-flash | 0.3% | -2 (0pts) |
| 37 | OpenAI: GPT-4o-mini (batch) | 0.3% | -2 (0pts) |
| 38 | OpenAI: GPT-5 Mini (batch) | 0.3% | -2 (0pts) |
| 39 | Anthropic: Claude Opus 4.7 (batch) | 0.3% | +8 (0pts) |
| 40 | OpenAI: GPT-5.4 (batch) | 0.3% | no change |
| 41 | Google: Gemini 3.5 Flash Lite (batch) | 0.3% | +1 (0pts) |
| 42 | Qwen: Qwen3.7 Flash New | 0.3% | - |
| 43 | MiniMax: MiniMax M2.7 | 0.3% | -5 (0pts) |
| 44 | OpenAI: gpt-oss-20b (free) | 0.3% | -5 (0pts) |
| 45 | Poolside: Laguna XS 2.1 (free) New | 0.2% | - |
| 46 | MoonshotAI: Kimi K2.6 | 0.2% | -5 (0pts) |
| 47 | OpenAI: GPT-5.5 (batch) | 0.2% | -2 (0pts) |
| 48 | OpenAI: GPT-5.4 Mini (batch) New | 0.2% | - |
This page is a permanent, unchanging record of real OpenRouter usage share as it stood on August 4, 2026. Later corrections to the data pipeline appear as new dated snapshots - this page's figures never change.
Source: OpenRouter (openrouter.ai/rankings), as of August 4, 2026.
Benchmark scores: Artificial Analysis (artificialanalysis.ai) via OpenRouter (openrouter.ai/rankings).
Methodology: smophy.ai/benchmark/methodology
Token counts originate from each provider's own tokenizer and are not directly comparable across providers.