Daily snapshot · immutable permalink
AI model usage rankings - September 14, 2026
As published on September 14, 2026, OpenAI: GPT-5.6 Luna (batch) led real OpenRouter usage share at 14.2%. Anthropic: Claude Opus 4.8 (batch), Anthropic: Claude Fable 5 (batch) entered the leaderboard for the first time. OpenAI: GPT-5.6 Terra (batch) moved 11 places up since September 13, 2026.
- #1 OpenAI: GPT-5.6 Luna (batch) - 14.2% usage share
- #2 DeepSeek: DeepSeek V4 Flash 0423 - 13.2% usage share
- #3 Tencent: Hy4 preview - 9.5% usage share
- #4 Z.ai: GLM 5.3 Flash (batch) - 8.7% usage share
- #5 DeepSeek: DeepSeek V4.1 Flash (batch) - 7.1% usage share
- #6 Xiaomi: MiMo-V2.5 - 6.4% usage share
- #7 Tencent: Hy3 - 4.3% usage share
- #8 NVIDIA: Nemotron 3 Ultra (free) - 2.5% usage share
- #9 DeepSeek: DeepSeek V4 Pro 0423 - 2.0% usage share
- #10 OpenAI: GPT-5.6 Sol (batch) - 1.7% usage share
| # | Model | Usage share | Rank change |
|---|---|---|---|
| 01 | OpenAI: GPT-5.6 Luna (batch) | 14.2% | no change |
| 02 | DeepSeek: DeepSeek V4 Flash 0423 | 13.2% | no change |
| 03 | Tencent: Hy4 preview | 9.5% | +1 (+1.3pts) |
| 04 | Z.ai: GLM 5.3 Flash (batch) | 8.7% | -1 (+0.3pts) |
| 05 | DeepSeek: DeepSeek V4.1 Flash (batch) | 7.1% | no change |
| 06 | Xiaomi: MiMo-V2.5 | 6.4% | no change |
| 07 | Tencent: Hy3 | 4.3% | +1 (+1.5pts) |
| 08 | NVIDIA: Nemotron 3 Ultra (free) | 2.5% | -1 (-0.3pts) |
| 09 | DeepSeek: DeepSeek V4 Pro 0423 | 2.0% | no change |
| 10 | OpenAI: GPT-5.6 Sol (batch) | 1.7% | +6 (+0.8pts) |
| 11 | Google: Gemini 3.8 Flash (batch) | 1.6% | +2 (+0.5pts) |
| 12 | Z.ai: GLM 5.3 (batch) | 1.6% | -2 (0pts) |
| 13 | Meta: Muse Spark 1.3 Contributor | 1.5% | -2 (0pts) |
| 14 | Anthropic: Claude Sonnet 5 (batch) | 1.3% | +6 (+0.6pts) |
| 15 | Z.ai: GLM 5.2 (free) | 1.3% | +3 (+0.4pts) |
| 16 | Upstage: Solar Pro 4 | 1.3% | -4 (-0.1pts) |
| 17 | MiniMax: MiniMax M3 | 1.2% | -3 (+0.1pts) |
| 18 | MoonshotAI: Kimi K3 (batch) | 1.1% | -3 (+0.1pts) |
| 19 | Anthropic: Claude Opus 5 (batch) | 0.9% | +5 (+0.4pts) |
| 20 | Poolside: Laguna S 2.1 (free) | 0.9% | -3 (0pts) |
| 21 | inclusionAI: Ling 3.0 Flash Fin (free) | 0.9% | -2 (+0.1pts) |
| 22 | Google: Gemini 3 Flash Preview (batch) | 0.7% | no change |
| 23 | Google: Gemini 2.5 Flash Lite (batch) | 0.6% | -2 (-0.1pts) |
| 24 | OpenAI: GPT-6 Astra (batch) | 0.6% | +4 (+0.2pts) |
| 25 | Google: Gemini 3.7 Flash (batch) | 0.5% | -2 (-0.1pts) |
| 26 | OpenAI: GPT-5.6 Terra (batch) | 0.5% | +11 (+0.2pts) |
| 27 | OpenAI: gpt-oss-120b | 0.5% | +4 (+0.1pts) |
| 28 | inclusionAI: Ling 3.0 Flash VL (free) | 0.5% | -1 (+0.1pts) |
| 29 | Google: Gemini 3.1 Flash Lite (batch) | 0.4% | +9 (+0.1pts) |
| 30 | Qwen: Qwen3.8 Flash | 0.4% | -5 (-0.1pts) |
| 31 | Google: Gemini 2.5 Flash (batch) | 0.4% | +5 (+0.1pts) |
| 32 | Anthropic: Claude Fable 5.1 (batch) | 0.4% | no change |
| 33 | Anthropic: Claude Sonnet 4.6 (batch) | 0.4% | +10 (+0.2pts) |
| 34 | NVIDIA: Nemotron 3.5 Lightning (free) | 0.4% | -8 (0pts) |
| 35 | DeepSeek: DeepSeek V3.2 | 0.3% | -6 (0pts) |
| 36 | SpaceXAI: Grok 4.6 | 0.3% | +9 (+0.1pts) |
| 37 | Anthropic: Claude Opus 4.8 (batch) New | 0.3% | - |
| 38 | Google: Gemma 4 31B (free) | 0.3% | +1 (0pts) |
| 39 | Thinking Machines: Inkling (free) | 0.3% | -5 (0pts) |
| 40 | Qwen: Qwen3.7 Flash | 0.3% | no change |
| 41 | Nex AGI: Nex-N2.5-Pro (free) | 0.3% | -11 (0pts) |
| 42 | NVIDIA: Nemotron 3 Super (free) | 0.3% | -9 (0pts) |
| 43 | Qwen: Qwen3.8 Max | 0.3% | -2 (0pts) |
| 44 | OpenAI: GPT-5.6 Luna Pro (batch) | 0.3% | +2 (+0.1pts) |
| 45 | Mistral: Mistral Nemo | 0.2% | -10 (-0.1pts) |
| 46 | Anthropic: Claude Fable 5 (batch) New | 0.2% | - |
This page is a permanent, unchanging record of real OpenRouter usage share as it stood on September 14, 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 September 14, 2026.
Benchmark scores: Artificial Analysis (artificialanalysis.ai) via OpenRouter (openrouter.ai/rankings).
Methodology: www.smophy.ai/benchmark/methodology
Token counts originate from each provider's own tokenizer and are not directly comparable across providers.
