Market share · 2026-09-12
LLM market share
Share of real API traffic, measured in tokens across public AI gateways. Not survey data, not vendor claims — the tokens that actually moved yesterday.
Leading lab
24.7%
OpenAI
Leading model
22.75%
OpenAI: GPT-5.6 Luna
Open-weight share
58.8%
of tracked tokens
Tokens / day
16T
50 models
The state of the LLM market in September 2026
OpenAI holds 24.7% of tracked tokens across 5 models, +14.20pp over 30 days. DeepSeek follows at 21.4%, then Tencent at 11.5%. Open-weight models together account for 58.8% of traffic — the single clearest signal of how much of the market is price-driven rather than brand-driven.
Token share and dollar share are different markets. The tracked models represent about $26.0M per day at list prices, and the labs at the top of the token table are rarely the labs at the top of the spend table. The spend view separates the two.
These figures are free to cite under CC BY 4.0. Attribute to “Tidelines (tidelines.ai),2026-09-12”.
Market share by lab
Share of gateway tokens on the latest archived day, all models combined.
| # | Lab | Share % | 30d Δ | Models | Top model |
|---|---|---|---|---|---|
| 1 | OpenAI | 24.72 | +14.20pp | 5 | OpenAI: GPT-5.6 Luna |
| 2 | DeepSeek | 21.44 | -8.84pp | 7 | DeepSeek: DeepSeek V4.1 Flash |
| 3 | Tencent | 11.54 | -4.32pp | 2 | Tencent: Hy4 preview |
| 4 | Z-AI | 10.28 | +2.13pp | 3 | Z.ai: GLM 5.3 Flash |
| 5 | Other | 6.83 | +1.70pp | 10 | Meta: Muse Spark 1.3 Contributor |
| 6 | Xiaomi | 5.63 | +1.46pp | 1 | Xiaomi: MiMo-V2.5 |
| 7 | Nvidia | 4.53 | +1.72pp | 3 | NVIDIA: Nemotron 3 Ultra |
| 8 | 3.91 | -1.58pp | 7 | Google: Gemini 3.8 Flash | |
| 9 | Anthropic | 2.09 | -4.19pp | 5 | Anthropic: Claude Sonnet 5 |
| 10 | MiniMax | 1.06 | -0.85pp | 1 | MiniMax: MiniMax M3 |
| 11 | Moonshot | 0.98 | -1.18pp | 1 | MoonshotAI: Kimi K3 |
| 12 | Alibaba | 0.95 | +0.33pp | 3 | Qwen: Qwen3.8 Max (0902) |
| 13 | Mistral | 0.25 | +0.03pp | 1 | Mistral: Mistral Nemo |
| 14 | xAI | 0.21 | -0.38pp | 1 | SpaceXAI: Grok 4.6 |
Market share by model
Top 20 individual models by share of gateway tokens.
| # | Model | Lab | Share % | 30d Δ |
|---|---|---|---|---|
| 1 | OpenAI: GPT-5.6 Luna | OpenAI | 22.75 | +15.66pp |
| 2 | Tencent: Hy4 preview | Tencent | 8.76 | +6.35pp |
| 3 | DeepSeek: DeepSeek V4.1 Flash | DeepSeek | 8.37 | +4.84pp |
| 4 | DeepSeek: DeepSeek V4 Flash 0731 | DeepSeek | 7.88 | -6.79pp |
| 5 | Z.ai: GLM 5.3 Flash | Z-AI | 7.88 | +5.92pp |
| 6 | Xiaomi: MiMo-V2.5 | Xiaomi | 5.63 | +1.46pp |
| 7 | NVIDIA: Nemotron 3 Ultra | Nvidia | 3.47 | +2.17pp |
| 8 | DeepSeek: DeepSeek V4 Flash 0423 | DeepSeek | 2.97 | -3.18pp |
| 9 | Tencent: Hy3 | Tencent | 2.78 | -10.67pp |
| 10 | Meta: Muse Spark 1.3 Contributor | Other | 1.50 | +0.74pp |
| 11 | Upstage: Solar Pro 4 | Other | 1.41 | +0.93pp |
| 12 | Z.ai: GLM 5.3 | Z-AI | 1.37 | +0.83pp |
| 13 | Google: Gemini 3.8 Flash | 1.34 | +1.04pp | |
| 14 | DeepSeek: DeepSeek V4 Pro 0423 | DeepSeek | 1.10 | -2.86pp |
| 15 | MiniMax: MiniMax M3 | MiniMax | 1.06 | -0.85pp |
| 16 | Z.ai: GLM 5.2 | Z-AI | 1.03 | -4.62pp |
| 17 | Poolside: Laguna S 2.1 | Other | 0.98 | -1.25pp |
| 18 | MoonshotAI: Kimi K3 | Moonshot | 0.98 | -1.18pp |
| 19 | Anthropic: Claude Sonnet 5 | Anthropic | 0.76 | -1.01pp |
| 20 | OpenAI: GPT-5.6 Sol | OpenAI | 0.76 | -0.19pp |
LLM market share FAQ
- Which AI lab has the largest LLM market share?
- OpenAI leads with 24.7% of tracked gateway tokens as of 2026-09-12, ahead of DeepSeek at 21.4%.
- How is LLM market share measured here?
- By tokens, not revenue or headlines. Tidelines archives daily token volumes published by public AI gateways (OpenRouter and Vercel AI Gateway) and computes each model's and lab's share of that total. It is a measure of developer API traffic, not of consumer chatbot users.
- How many models and labs are tracked?
- 50 models across 14 labs on the latest archived day (2026-09-12), representing roughly 16T tokens per day.
- Can I cite these market share numbers?
- Yes. The data is published under CC BY 4.0 — cite "Tidelines (tidelines.ai)" with the date of the snapshot you used.