35.1×
more tokens a week than a year ago: 193.6T now vs 5.5T.
NEWSllama.cpp b12027
The AI model market, today
Demand, momentum, benchmarks and what people are saying, refreshed from public sources.
Demand
Daily tokens by model. Hover a day to see every model.
Last 7 days181.4T tokens
vs previous 7 days▲ 3.9%
1 Nov vs day before▼ 7.0%
tokens per day
| Day | DeepSeek V4.1 Flash 1017 | GLM 5.3 Flash | DeepSeek V4.1 Flash | GPT-6 Luna | Hy4 preview | Claude Opus 5.5 | Step 5 Preview | MiniMax M3.1 Flash | Top 50 models |
|---|---|---|---|---|---|---|---|---|---|
| 26 October 2026 | 5.4T tokens | 2.3T tokens | 2.2T tokens | 1.5T tokens | 1.4T tokens | 1.0T tokens | 801.0B tokens | 678.2B tokens | 24.4T tokens |
| 27 October 2026 | 6.0T tokens | 1.9T tokens | 2.4T tokens | 1.1T tokens | 1.4T tokens | 1.1T tokens | 801.7B tokens | 844.3B tokens | 25.1T tokens |
| 28 October 2026 | 7.3T tokens | 2.4T tokens | 2.4T tokens | 1.4T tokens | 1.5T tokens | 1.0T tokens | 960.2B tokens | 899.1B tokens | 27.6T tokens |
| 29 October 2026 | 8.1T tokens | 2.5T tokens | 2.2T tokens | 2.8T tokens | 1.7T tokens | 1.3T tokens | 1.2T tokens | 966.3B tokens | 31.2T tokens |
| 30 October 2026 | 8.2T tokens | 2.7T tokens | 1.8T tokens | 1.2T tokens | 1.5T tokens | 993.5B tokens | 959.9B tokens | 1.1T tokens | 28.5T tokens |
| 31 October 2026 | 7.1T tokens | 1.8T tokens | 1.7T tokens | 810.4B tokens | 958.9B tokens | 620.9B tokens | 799.7B tokens | 891.9B tokens | 23.1T tokens |
| 1 November 2026 | 6.0T tokens | 1.4T tokens | 2.1T tokens | 802.9B tokens | 734.6B tokens | 654.1B tokens | 747.8B tokens | 900.3B tokens | 21.4T tokens |
The 7-day view covers the top 50 models, day by day. The 8 biggest models are drawn; smaller ones are left out of the stacks. Headline numbers and the tooltip total count everything.
DeepSeek V4 Flash
DeepSeek · 4 models in top 100
69.6T
tokens / week
▲ 19% vs last week
GLM 5 Flash
Z.ai · 1 model in top 100
15.0T
tokens / week
▲ 13% vs last week
GPT-6 Luna
OpenAI · 1 model in top 100
9.7T
tokens / week
▲ 15% vs last week
Hy4
Tencent · 1 model in top 100
9.1T
tokens / week
▲ 1.1% vs last week
Claude Opus 5
Anthropic · 2 models in top 100
7.0T
tokens / week
▼ 23% vs last week
S
Step 5
StepFun · 1 model in top 100
6.2T
tokens / week
▲ 72% vs last week
MiniMax M3 Flash
MiniMax · 1 model in top 100
6.2T
tokens / week
▲ 690% vs last week
GLM 5
Z.ai · 2 models in top 100
5.1T
tokens / week
▼ 15% vs last week
T
JEV 1
Typesafe · 2 models in top 100
4.9T
tokens / week
▲ 6.5% vs last week
GPT-5 Luna
OpenAI · 1 model in top 100
4.4T
tokens / week
▲ 154% vs last week
Families of the top 100 models; daily tokens over 14 days.
Trending models
Why each model is here, with the number behind it
| # | Model | Why trending | Tokens / week | 14 days | Best for | Context · list price | LiveBench |
|---|---|---|---|---|---|---|---|
| 1 | New this month | — | — | — | — — | — | |
| 2 | New this month | — | — | — | — — | — | |
| 3 | New this month | — | — | — | — — | — | |
| 4 | Gemini 4 Carbon Google | New this month506 mentions | — | — | — | — — | — |
| 5 | Mercury 2.5 Preview Inception | New this month | — | — | — | 254K — | — |
| 6 | Kimi K3 Moonshot AI · Open weights | 234 mentionsDemand #24 | 1.7T ▼ 14% | Reasoning · #9 | 1M $0.19 / $15.00 | 64.6 | |
| 7 | DeepSeek V4.1 Flash DeepSeek · Open weights | Demand #3165 mentionsNew this quarter | 14.9T ▼ 17% | Vision tasks | 1M $0.04 / $0.50 | — | |
| 8 | MiniMax M3.1 Flash MiniMax · Open weights | +690% usageDemand #844 mentions | 6.2T ▲ 690% | Browser · #8 | 1M $0.15 / $0.60 | 66.9 | |
| 9 | DeepSeek V3.1 DeepSeek · Open weights | — | 49.4B ▲ 1.3% | — | — | 160K $0.25 / $0.95 | — |
| 10 | Claude Opus 5.5 Anthropic | 526 mentionsDemand #6New this quarter | 6.7T ▼ 19% | Design · #1 | 1M $4.00 / $20.00 | 85.1 | |
| 11 | GPT-5.6 Luna OpenAI | +154% usageDemand #95 mentions | 4.4T ▲ 154% | Vision tasks | 1M $0.20 / $1.20 | — | |
| 12 | DeepSeek V4 Flash 0731 DeepSeek · Open weights | Demand #11 | 3.9T ▼ 1.3% | — | 1M $0.12 / $0.24 | — | |
| 13 | Gemma 3 27B Google · Open weights | — | 42.1B ▲ 6.2% | — | Vision tasks | 128K $0.08 / $0.35 | — |
| 14 | Qwen3 VL 235B A22B Alibaba · Open weights | — | 28.7B ▲ 29% | — | Vision tasks | 256K $0.21 / $1.90 | — |
| 15 | GLM 5.3 Z.ai · Open weights | 110 mentionsDemand #10New this quarter | 4.2T ▼ 16% | — | 1M $0.08 / $3.60 | — | |
| 16 | S Step 5 Preview StepFun · Open weights | Demand #7+72% usage3 mentions | 6.2T ▲ 72% | Coding · #16 | 1M $1.00 / $2.70 | 58.6 | |
| 17 | Claude Opus 5 Anthropic | 117 mentionsDemand #65 | 228.1B ▼ 68% | — | Vision tasks | 1M $5.00 / $25.00 | — |
| 18 | DeepSeek V4.1 Flash 1017 DeepSeek · Open weights | Demand #1Trending with developers53 mentions | 48.1T ▲ 43% | Coding · #3 | 1M $0.15 / $0.60 | 72.9 | |
| 19 | Kimi K3.1 Pro Moonshot AI | 328 mentions+264% usageDemand #22 | 2.3T ▲ 264% | Reasoning · #8 | 1M $2.00 / $10.00 | 71.3 | |
| 20 | GPT-5 OpenAI | 53 mentions | 8.6B ▼ 16% | — | Vision tasks | 391K $1.25 / $10.00 | — |
Under the context window: the list price per 1M input / output tokens. Best for = the model's strongest LiveBench task type, with its rank.
Latest news & research
Official posts, papers and release notes
- Community · thenewstack.ioGPT-6.1 Terra vs. Claude Opus 5.5: Cheaper and faster in three tests at max2 Nov · about Claude Opus 5.5, GPT-6.1 Terra
- Community · fork.zoneChinese labs now target OpenAI's GPT-6 Astra for distillation1 Nov · about GPT-6 Astra
- Community · docs.mistral.aiTell HN: Mistrap Pro includes DeepSeek v4.1 Flash 10171 Nov · about DeepSeek V4.1 Flash 1017
- Release note · llama.cppllama.cpp b120271 Nov · about Claude Opus 5.5
- Community · github.comMemdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard31 Oct · about Qwen3.8 27B
- Community · simonwillison.netGPT-6.1 Terra31 Oct · about GPT-6.1 Terra
- Community · ryanmadden.netGPT-5.6 Sol can beat Factorio on peaceful mode31 Oct · about GPT-5.6 Sol
- Community · twitter.comQwen3.8-Flash Now Available in the EU on Alibaba Cloud31 Oct · about Qwen3.8 Flash
- Release note · llama.cppllama.cpp b1201931 Oct · about Qwen 3 Coder 480b, Claude Opus 5.5
- Community · skypilot.aiPost-training 4B model with RL to fly Tesla Roadster 21% better than GPT-6 Astra30 Oct · about GPT-6 Astra
- Community · scraping.clubHermes and REA and GLM 5.3 = Reverse engineering TikTok app for less than $2029 Oct · about GLM 5.3
- Research · arXivOn Zeros and Algorithms for Disordered Systems II: The Antiferromagnetic Ising Model up to Kesten-Stigum29 Oct · about GPT-5.5 Pro
- Research · arXivPre-Crime for Agents, at the Last Second: Probing When a Recruited Agent Gives In29 Oct · about Qwen3 32B, GPT OSS 20B
- Research · arXivIndicIPR-QA: A Benchmark for Extractive Question Answering on Indian IPR Laws29 Oct · about GPT-4o
- Engineering · AWSAccelerate agentic AI with NVIDIA Nemotron models on Amazon EKS and SageMaker HyperPod EKS29 Oct · about Nemotron 3.5 Super VL
- Release note · Hugging Facetransformers v5.22.0: v5.22.0: Mistral Large 4 (Le Chonk)29 Oct · about Venice Medium
- Community · reuters.comDeepSeek, Moonshot used OpenAI's GPT-6 Astra to train latest models29 Oct · about GPT-6 Astra
- Community · artificialanalysis.aiGemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v529 Oct · about Gemini 4 Carbon
- Official · Mistral AILe Chonk: Built for Coders, Trusted by Defenders.29 Oct · about Venice Medium
- Release note · llama.cppllama.cpp b1197629 Oct · about Claude Opus 5.5
By the numbers
Facts computed from this week's data, each with its source
71%
of tokens go to open-weight models, up from 28% a year ago.
Open 71% · Closed 29%
37%
of this week's tokens went to DeepSeek, the biggest vendor.
DeepSeek 37%, OpenAI 12%, Z.ai 11%, Anthropic 7%, Tencent 6%, MiniMax 4% of this week's tokens.
6.2T
tokens in a week for MiniMax M3.1 Flash, the model of the week.
+690%
week-over-week growth for MiniMax M3.1 Flash (MiniMax), the fastest climber.
10.6T
tokens a week used by Claude Code, the biggest app.
85.7
overall LiveBench score for Claude Fable 5.1, the current leader.
$0.18
per successful task for DeepSeek V4.1 Flash 1017, the cheapest of LiveBench's top 10.
Benchmarks
LiveBench release 25 Oct 2026. Pick a task type to rank by it.
- 1st · OverallClaude Fable 5.1Anthropic · max effort85.7/ 100
- best in Overall
- best in Long Agentic
- best in Coding
- 2nd · OverallGPT-6 AstraOpenAI · max effort85.6/ 100
- 3rd · OverallClaude Opus 5.5Anthropic · max effort85.1/ 100
- best in Design
| # | Model | Overall | Long Agentic | Reasoning | Coding | Browser | Data Analysis | Research | Design | Cost per successful task |
|---|---|---|---|---|---|---|---|---|---|---|
| 4 | GPT-6.1 Sol OpenAI · max effort | 84.1 | Long Agentic 68.7 | Reasoning 92.6 | Coding 96.0 | Browser 93.3, best in column | Data Analysis 82.7 | Research 83.5 | Design 82.1 | $0.82 |
| 5 | Claude Sonnet 5.5 Anthropic · max effort | 80.0 | Long Agentic 63.4 | Reasoning 91.6 | Coding 94.7 | Browser 89.3 | Data Analysis 59.5 | Research 83.6 | Design 85.0 | $4.14 |
| 6 | Gemini 4 Argon Google · high effort | 79.3 | Long Agentic 61.2 | Reasoning 95.1, best in column | Coding 89.7 | Browser 82.8 | Data Analysis 85.2, best in column | Research 83.1 | Design 73.7 | $1.92 |
| 7 | GPT-6.1 Terra OpenAI · max effort | 77.8 | Long Agentic 48.6 | Reasoning 88.7 | Coding 94.3 | Browser 93.1 | Data Analysis 82.3 | Research 73.9 | Design 82.3 | $0.67 |
| 8 | Muse Wonder 1.0 Meta · max effort | 74.2 | Long Agentic 52.6 | Reasoning 90.6 | Coding 86.2 | Browser 60.0 | Data Analysis 81.6 | Research 88.0, best in column | Design 75.1 | $2.01 |
| 9 | DeepSeek V4.1 Flash 1017 DeepSeek · max effort | 72.9 | Long Agentic 35.7 | Reasoning 89.1 | Coding 96.0 | Browser 67.2 | Data Analysis 79.5 | Research 79.5 | Design 83.4 | $0.18 |
| 10 | Kimi K3.1 Pro Moonshot AI · max effort | 71.3 | Long Agentic 39.1 | Reasoning 91.1 | Coding 81.8 | Browser 75.9 | Data Analysis 79.1 | Research 77.9 | Design 77.8 | $1.04 |
| 11 | Grok 4.7 xAI · xhigh effort | 70.1 | Long Agentic 46.1 | Reasoning 82.7 | Coding 84.5 | Browser 49.3 | Data Analysis 76.9 | Research 86.4 | Design 78.0 | $2.91 |
| 12 | Claude Haiku 5.5 Anthropic · max effort | 69.8 | Long Agentic 40.0 | Reasoning 82.4 | Coding 91.4 | Browser 69.5 | Data Analysis 53.1 | Research 81.9 | Design 83.1 | $0.91 |
| 13 | MiMo V2.6 Pro Xiaomi · high effort | 68.5 | Long Agentic 35.2 | Reasoning 79.6 | Coding 81.5 | Browser 77.3 | Data Analysis 72.3 | Research 76.7 | Design 78.5 | $0.18 |
| 14 | GPT-5.6 Terra OpenAI · max effort | 67.8 | Long Agentic 33.1 | Reasoning 90.6 | Coding 86.2 | Browser 70.3 | Data Analysis 79.3 | Research 60.4 | Design 76.0 | $1.92 |
| 15 | Grok 4.6 xAI · xhigh effort | 67.8 | Long Agentic 28.0 | Reasoning 91.2 | Coding 89.1 | Browser 49.4 | Data Analysis 75.3 | Research 84.5 | Design 79.8 | $1.42 |
| 16 | MiniMax M3.1 Flash MiniMax · max effort | 66.9 | Long Agentic 28.8 | Reasoning 82.9 | Coding 83.9 | Browser 80.6 | Data Analysis 71.4 | Research 72.7 | Design 74.0 | $0.15 |
| 17 | GPT-6 Luna OpenAI · max effort | 64.6 | Long Agentic 30.9 | Reasoning 81.8 | Coding 84.5 | Browser 70.3 | Data Analysis 73.4 | Research 43.5 | Design 82.5 | $0.12 |
| 18 | Kimi K3 Moonshot AI · max effort | 64.6 | Long Agentic 26.4 | Reasoning 90.7 | Coding 59.3 | Browser 75.2 | Data Analysis 78.7 | Research 79.0 | Design 77.2 | $2.09 |
Cell colour deepens with the score; the number is always shown. ★ marks the best in each column. Hover a cell, or focus it and use the arrow keys, for its sub-areas.
Model universe
Vendors, families and models, sized by weekly tokens. Drag nodes, Ctrl/⌘ + scroll or pinch to zoom, click to inspect.
Top 50 models this week, by vendor
- DeepSeek: 71.0T tokens this week
- DeepSeek V4.1 Flash 1017, #1, 48.1T tokens
- DeepSeek V4.1 Flash, #3, 14.9T tokens
- DeepSeek V4 Flash 0731, #11, 3.9T tokens
- DeepSeek V4 Flash, #18, 2.7T tokens
- DeepSeek V4 Pro 0813, #31, 794.3B tokens
- DeepSeek V4 Pro, #41, 573.1B tokens
- Z.ai: 20.2T tokens this week
- GLM 5.3 Flash, #2, 15.0T tokens
- GLM 5.3, #10, 4.2T tokens
- GLM 5.2, #30, 920.8B tokens
- OpenAI: 19.9T tokens this week
- GPT-6 Luna, #4, 9.7T tokens
- GPT-5.6 Luna, #9, 4.4T tokens
- GPT-6.1 Sol, #12, 3.8T tokens
- GPT-6.1 Terra, #34, 645.7B tokens
- GPT-6 Luna Pro, #38, 609.3B tokens
- GPT-6 Astra, #45, 444.7B tokens
- GPT OSS 120B, #49, 385.8B tokens
- Anthropic: 12.7T tokens this week
- Claude Opus 5.5, #6, 6.7T tokens
- Claude Haiku 5.5, #17, 2.8T tokens
- Claude Sonnet 5.5, #19, 2.6T tokens
- Claude Sonnet 5, #42, 516.6B tokens
- Tencent: 12.1T tokens this week
- Hy4 preview, #5, 9.1T tokens
- Tencent HY3, #14, 3.1T tokens
- MiniMax: 6.8T tokens this week
- MiniMax M3.1 Flash, #8, 6.2T tokens
- MiniMax M3, #37, 612.2B tokens
- StepFun: 6.7T tokens this week
- Step 5 Preview, #7, 6.2T tokens
- Step 3.7 Flash, #43, 499.9B tokens
- NVIDIA: 6.6T tokens this week
- Nemotron 3 Ultra, #15, 3.0T tokens
- Nemotron 3.5 Super VL, #21, 2.6T tokens
- Nemotron 3.5 Lightning, #29, 1.0T tokens
- Typesafe: 4.9T tokens this week
- JEV 1.13 (TypeSafe), #13, 3.1T tokens
- JEV 1.14 (TypeSafe), #23, 1.7T tokens
- Google: 4.8T tokens this week
- Gemini 3.8 Flash, #25, 1.5T tokens
- Gemini 4 Argon, #28, 1.2T tokens
- Gemini 3.1 Flash Lite, #35, 623.0B tokens
- Gemini 3 Flash Preview, #36, 620.5B tokens
- Gemma 4 31B IT, #44, 494.6B tokens
- Gemini 3.7 Flash, #48, 393.8B tokens
- Xiaomi: 4.3T tokens this week
- MiMo-V2.6-Flash, #16, 3.0T tokens
- MiMo-V2.6-Pro, #27, 1.3T tokens
- Moonshot AI: 4.0T tokens this week
- Kimi K3.1 Pro, #22, 2.3T tokens
- Kimi K3, #24, 1.7T tokens
- Meta: 2.6T tokens this week
- Muse Spark 1.3 Contributor, #20, 2.6T tokens
- Upstage: 1.8T tokens this week
- Solar Mini 4, #26, 1.4T tokens
- Solar Pro 4, #47, 396.8B tokens
- Alibaba: 1.6T tokens this week
- Qwen3.8 Flash, #39, 600.1B tokens
- Qwen3.8 27B, #40, 598.4B tokens
- Qwen3.7 Flash, #46, 441.5B tokens
- Poolside: 735.0B tokens this week
- Laguna S 2.1, #32, 735.0B tokens
- Dots Studio: 731.6B tokens this week
- Dots3-Note Preview, #33, 731.6B tokens
- Thinking Machines: 377.9B tokens this week
- Inkling, #50, 377.9B tokens
Node size: weekly tokens. A family node appears when a family has two or more models in view.
Trends
How the frontier moved, and what each point of intelligence costs
The best Intelligence Index score each lab had released by each date
Intelligence Index
- Google59.4
- Anthropic58.4
- OpenAI56.9
- Meta55.8
- Moonshot AI55.1
- xAI46.4
- Xiaomi46.3
- Alibaba45.4
Media leaderboards
Top models by Elo from blind side-by-side votes
| # | Model | Elo | 95% CI | Released |
|---|---|---|---|---|
| 1 | 1,222 | ±10 | Sep 2026 | |
| 2 | 1,207 | ±10 | Sep 2026 | |
| 3 | GPT Image 2 (high)OpenAI | 1,186 | ±11 | Apr 2026 |
| 4 | 1,176 | ±12 | Oct 2026 | |
| 5 | MMAI-Image-2.6Microsoft AI | 1,166 | ±12 | Aug 2026 |
| 6 | SGrok Imagine Image 2.0SpaceXAI | 1,166 | ±12 | Aug 2026 |
| 7 | Nano Banana 2 (Gemini 3.1 Flash Image)Google | 1,140 | ±10 | Feb 2026 |
| 8 | Muse ImageMeta | 1,127 | ±11 | Jul 2026 |
| 9 | GPT Image 1.5 (high)OpenAI | 1,123 | ±11 | Dec 2025 |
| 10 | MMAI-Image-2.6-FlashMicrosoft AI | 1,119 | ±12 | Sep 2026 |
What people are watching and discussing
The last two weeks, ranked by views and engagement
- AI finally sleeps, Gemini 4 Carbon, ETH cracked, energy-based models, Anthropic IPO: AI NEWSAI Search · 250.6K views · 1 Nov
- GPT-6.1 Terra is now in orbitOpenAI · 517.8K views · 26 Oct
- Gemini 4 Carbon: The New FrontierPrompt Engineering · 351.1K views · 28 Oct
- Запуск ИИ модели GLM 5.2 без видеокарты на ноутбукеВлад Нейронов🤖 · 303.1K views · 26 Oct
- Люди начали подключать GPT-6 Astra ко всему подряд: 9 примеров, от которых инженеры в шокеDenTrave · 56.2K views · 1 Nov
- HUGE GPT-6.2 BEL LEAK! Opus 6, Gemini 4 Carbon First Look, Claude Fable 5.5, Qwen 4 & More! AI NEWSWorldofAI · 129.1K views · 30 Oct
- HUGE ANTHROPIC BOMBSHELL! Claude REA UNBANNED?! GPT-6.1 Terra, Qwen 4 27B, & Fable 5.5! AI NEWSWorldofAI · 157.0K views · 27 Oct
- Kimi K3.1 Pro, el modelo abierto... ¿al nivel de Opus?midudev · 127.5K views · 28 Oct
- Kimi K3.1 против Opus 5,5 и GPT-6 Astra – Я ВСЕ ПРОВЕРИЛПродуктивный Совет · 88.5K views · 29 Oct
- Claude 2.0 es una locura (todas las novedades explicadas)Xavier Mitjana · 101.2K views · 28 Oct
- OpenArt Nano Banana 2.1 with Seedance 2.5 – Full AI Product Campaign WorkflowEigi and AI · 129.8K views · 25 Oct
- Every Claude 2.0 Concept Explained for Non-CodersNate Herk | AI Automation · 53.9K views · 30 Oct
Latest AI research
New papers from arXiv and the models they discuss
- Algebraic World Model: From Exploration to ProofsarXiv · 29 Oct
- Spectral isometries of unital $C^*$-algebrasarXiv · 29 Oct
- No tracked model named
- On separators in $K_t$-minor-free graphsarXiv · 29 Oct
Demand, benchmarks, news, research and social posts are refreshed from public sources. List prices are public prices per 1M tokens.