Leaderboard · 2026-09-13

The best LLMs in 2026

An LLM leaderboard built on real usage. Every model here is ranked by the tokens it actually processed on public AI gateways — the closest thing to a revealed preference for which models are worth running in production.

How "best" is measured in September 2026

50 models carried measurable traffic on the latest archived day, together moving about 16T tokens across 14 labs. Rather than scoring models on test sets, this page reads three signals off that traffic: raw share (who wins the most tokens), 30-day momentum (who is being adopted), and usage per dollar of list price (who is worth the money).

The current leader is OpenAI: GPT-5.6 Luna at 23.47% of tracked tokens, +16.98pp over 30 days. Because each day is archived, every change figure compares two real days rather than a rolling estimate.

Top 25 LLMs by real usage

Share of gateway tokens on the latest archived day.

Tidelines — LLM usage and market share
#ModelLabShare %Tokens/d30d Δ
1OpenAI: GPT-5.6 LunaOpenAI23.473.9T+16.98pp
2DeepSeek: DeepSeek V4 Flash 0731DeepSeek8.621.4T-5.47pp
3Z.ai: GLM 5.3 FlashZ-AI8.421.4T+6.46pp
4Tencent: Hy4 previewTencent8.251.4T+5.84pp
5DeepSeek: DeepSeek V4.1 FlashDeepSeek7.741.3T+4.20pp
6Xiaomi: MiMo-V2.5Xiaomi5.89985B+1.81pp
7DeepSeek: DeepSeek V4 Flash 0423DeepSeek3.10519B-2.51pp
8NVIDIA: Nemotron 3 UltraNvidia2.81470B-0.28pp
9Tencent: Hy3Tencent2.74458B-9.61pp
10Z.ai: GLM 5.3Z-AI1.65277B+1.11pp
11Meta: Muse Spark 1.3 ContributorOther1.53256B+0.77pp
12Upstage: Solar Pro 4Other1.36227B+0.92pp
13Google: Gemini 3.8 FlashGoogle1.20200B+0.90pp
14DeepSeek: DeepSeek V4 Pro 0423DeepSeek1.13188B-2.03pp
15MiniMax: MiniMax M3MiniMax1.05175B-0.91pp
16MoonshotAI: Kimi K3Moonshot1.00167B-0.90pp
17OpenAI: GPT-5.6 SolOpenAI0.96161B+0.12pp
18Poolside: Laguna S 2.1Other0.95160B-1.15pp
19Z.ai: GLM 5.2Z-AI0.93156B-6.93pp
20inclusionAI: Ling 3.0 Flash FinOther0.80133B+0.51pp
21Anthropic: Claude Sonnet 5Anthropic0.79133B-0.68pp
22Google: Gemini 2.5 Flash LiteGoogle0.66110B-0.23pp
23Google: Gemini 3 Flash PreviewGoogle0.61102B-0.40pp
24Google: Gemini 3.7 FlashGoogle0.60101B-0.42pp
25Dots Studio: Dots3-Note Preview (free)Other0.5999B+0.40pp
Data: OpenRouter as of 2026-09-13 · Source: tidelines.ai/best-llms-2026

Best value: usage per dollar

Token share divided by output list price per million.

Tidelines — LLM usage and market share
ModelShare %$/M outValue
DeepSeek: DeepSeek V4 Flash 07318.62$0.1271.81
Xiaomi: MiMo-V2.55.89$0.2821.04
OpenAI: GPT-5.6 Luna23.47$1.2019.56
DeepSeek: DeepSeek V4 Flash 04233.10$0.1817.51
Z.ai: GLM 5.3 Flash8.42$0.5016.83
DeepSeek: DeepSeek V4.1 Flash7.74$0.6012.90
Mistral: Mistral Nemo0.31$0.0310.32
Meta: Muse Spark 1.3 Contributor1.53$0.207.64
Data: OpenRouter as of 2026-09-13 · Source: tidelines.ai/best-llms-2026

Biggest movers over 30 days

Change in share of gateway tokens, percentage points.

Tidelines — LLM usage and market share
ModelLabShare %30d Δ
OpenAI: GPT-5.6 LunaOpenAI23.47+16.98pp
Z.ai: GLM 5.3 FlashZ-AI8.42+6.46pp
Tencent: Hy4 previewTencent8.25+5.84pp
DeepSeek: DeepSeek V4.1 FlashDeepSeek7.74+4.20pp
Xiaomi: MiMo-V2.5Xiaomi5.89+1.81pp
Z.ai: GLM 5.3Z-AI1.65+1.11pp
Data: OpenRouter as of 2026-09-13 · Source: tidelines.ai/best-llms-2026

Best open-weight models

Open-weight releases now hold a meaningful slice of gateway traffic. The leaders by measured usage are DeepSeek: DeepSeek V4 Flash 0731, Z.ai: GLM 5.3 Flash, DeepSeek: DeepSeek V4.1 Flash. The open-weights index tracks that share over time.

Best LLMs in 2026 FAQ

What is the best LLM in 2026?
By the only measure that is not self-reported — how much real traffic a model carries — OpenAI: GPT-5.6 Luna from OpenAI leads in September 2026 with 23.47% of tokens routed through the public AI gateways Tidelines tracks.
Which LLMs are developers actually using in 2026?
The current top three by measured token volume are OpenAI: GPT-5.6 Luna, DeepSeek: DeepSeek V4 Flash 0731, Z.ai: GLM 5.3 Flash. In total 50 models registered real traffic on 2026-09-13, but usage is heavily concentrated in the top ten.
Which LLM gives the best value for money?
DeepSeek: DeepSeek V4 Flash 0731 (DeepSeek) currently has the strongest usage-per-dollar profile: 8.62% of tracked tokens at $0.12 per million output tokens.
Which model is growing fastest?
OpenAI: GPT-5.6 Luna has gained the most share over the last 30 days, at +16.98pp.
Is this a benchmark leaderboard?
No. Benchmarks measure how a model scores on test sets. This page measures what developers pay for and route production traffic to, computed daily from public gateway token counts.