Signal
Momentum across the AI ecosystem, computed daily from real data — how these numbers are calculated.
7 of 17 planned dashboards are built so far (4 as time-series tiles above, 3 as Observatory ranked lists below) — see content/information-architecture.md for the rest and the four dashboard-page templates they fall into (time-series, leaderboard, observatory, relationship map).
Hugging Face Observatory
2026-11-13 → 2026-11-14 — Edge0's 145-billion-parameter model, which the company says runs on a single gaming PC, more than doubled its downloads in a day (+139.3%), while copies of MiniMax's H3 image model kept spreading and Zyphra's ZONOS3-Mini speech model rose 51.1%.
| Repo / model | Activity | Change |
|---|---|---|
| Edge0/Edge0-145B-NVFP4 | 33,819 downloads | +139.3% |
| Abiray/MiniMax-H3-Image-GGUF | 111,241 downloads | +112.9% |
| Zyphra/ZONOS3-Mini | 19,743 downloads | +51.1% |
| HAL-X-AI/THX-02 | 17,171 downloads | +38.2% |
| Cactus-Compute/mirage | 24,452 downloads | +29.0% |
Source: ai-tracker, reports/2026-11-14.md (Hugging Face — Models, "Fastest-growing by downloads"; "uncensored"/"abliterated" variants left out under the editorial sensitivity rule)
OpenRouter Observatory
Day of 2026-11-13 — Baidu's free ERNIE 5.5 Flash Preview more than doubled again (+117.9%, to 920 billion tokens), Google's Gemini 4 Carbon rose 51.7%, and DeepSeek V4.1 Flash still carried 27.4% of the day's 207.6 trillion tokens across 681 models.
| Repo / model | Activity | Change |
|---|---|---|
| baidu/ernie-5.5-flash-preview-20261110 | 920.4B tokens | +117.9% |
| google/gemini-4-carbon-20261110 | 354.4B tokens | +51.7% |
| meta/muse-ember-20261109 | 360.7B tokens | +45.4% |
| anthropic/claude-fable-5.2-20261109 | 658.2B tokens | +44.1% |
| thinkingmachines/inkling-2-small-20261107 | 712.0B tokens | +40.5% |
Source: ai-tracker, reports/2026-11-14.md (OpenRouter — model usage, "Fastest-growing by tokens" >=50B base, top 5; totals and share from the same report's usage table)
Vercel AI Gateway Observatory
Latest daily shares as of Nov 3, 2026 (models: token volume, requests, spend), Nov 2 (labs) and Oct 29 (reach, preference); providers and apps ranked for the current week; Production Index covers September 2026 — Who uses the most tokens and who gets paid are two different leaderboards on Vercel's AI Gateway. Low-cost models like Beam 1.0 and DeepSeek's V4.1 Flash line handle the most volume, and TypeSafe AI's Jev carries more than a fifth of all requests. On spend, though, Anthropic's Claude Opus 5.5 led with 23.3% on Nov 3, and Anthropic took 35.2% of lab spend, far more than its 10.6% share of tokens.
| Repo / model | Activity |
|---|---|
| Beam 1.0 | 15.4% |
| DeepSeek V4.1 Flash 1017 | 14.2% |
| GLM 5.3 Flash | 8.7% |
| DeepSeek V4.1 Flash | 8.0% |
| GPT-6 Luna | 7.0% |
| Claude Opus 5.5 | 6.9% |
| GLM 5.3 | 6.1% |
| Grok 4.7 | 5.4% |
| Jev | 3.5% |
| Other (10 models) | 24.9% |
Share of gateway requests, Nov 3
| Model | Share |
|---|---|
| Jev | 21.7% |
| GPT-6 Luna | 9.7% |
| Jev 1.14.0 | 5.9% |
| DeepSeek V4.1 Flash 1017 | 4.4% |
| DeepSeek V4.1 Flash | 4.3% |
| GLM 5.3 Flash | 4.2% |
| Jev 1.13.0 | 3.3% |
| GLM 5.3 | 2.4% |
| Claude Haiku 5.5 | 2.2% |
| Other (10 models) | 42.0% |
Share of gateway spend, Nov 3
| Model | Share |
|---|---|
| Claude Opus 5.5 | 23.3% |
| Grok 4.7 | 16.3% |
| GLM 5.3 | 6.9% |
| Kimi K3 | 6.8% |
| Claude Sonnet 5.5 | 6.3% |
| GPT-6.1 Sol | 4.4% |
| GPT-6 Astra | 3.9% |
| Gemini 3.8 Flash | 2.1% |
| GLM 5.3 Flash | 1.9% |
| Other (10 models) | 28.1% |
Share of gateway teams using each model, Oct 29. Teams use several models, so shares don't sum to 100%.
| Model | Teams |
|---|---|
| Jev | 22.3% |
| Claude Sonnet 5.5 | 11.6% |
| GPT-6 Luna | 11.1% |
| Claude Haiku 5.5 | 11.0% |
| Claude Haiku 4.5 | 10.4% |
| text-embedding-3-small | 8.8% |
| Claude Opus 5.5 | 7.1% |
| Gemini 3.8 Flash | 6.9% |
| Claude Sonnet 5 | 6.7% |
| GPT 5.6 Luna | 6.3% |
Share of teams whose primary model (by token volume) it is, Oct 29
| Model | Teams |
|---|---|
| Jev | 15.4% |
| Claude Sonnet 5.5 | 5.7% |
| GPT-6 Luna | 5.5% |
| Claude Haiku 5.5 | 5.0% |
| Claude Haiku 4.5 | 4.3% |
| Claude Sonnet 5 | 3.3% |
| Claude Opus 5.5 | 3.2% |
| Gemini 2.5 Flash | 3.2% |
| GPT 5.6 Luna | 2.6% |
| Other (10 models) | 51.7% |
Share of gateway tokens, Nov 2
| Lab | Share |
|---|---|
| DeepSeek | 30.7% |
| Reflection | 14.7% |
| Z.AI | 14.3% |
| Anthropic | 10.6% |
| OpenAI | 10.2% |
| SpaceXAI | 6.6% |
| TypeSafe AI | 3.8% |
| 2.4% | |
| Stealth | 1.8% |
| Other (10 labs) | 4.9% |
Share of gateway spend, Nov 2
| Lab | Share |
|---|---|
| Anthropic | 35.2% |
| SpaceXAI | 23.8% |
| OpenAI | 13.3% |
| Z.AI | 8.6% |
| 7.0% | |
| DeepSeek | 5.6% |
| Moonshot AI | 4.8% |
| TypeSafe AI | 0.7% |
| Alibaba Cloud | 0.4% |
| Other (10 labs) | 0.7% |
Share of gateway requests, Nov 2
| Lab | Share |
|---|---|
| TypeSafe AI | 29.6% |
| OpenAI | 20.2% |
| DeepSeek | 14.2% |
| 9.8% | |
| Anthropic | 7.0% |
| Z.AI | 6.8% |
| Alibaba Cloud | 2.5% |
| SpaceXAI | 2.4% |
| Reflection | 1.9% |
| Other (10 labs) | 5.6% |
By token volume. Vercel publishes the ranking order only, not shares.
| Provider | Rank |
|---|---|
| Reflection | #1 |
| Baseten | #2 |
| Fireworks | #3 |
| DeepSeek | #4 |
| OpenAI | #5 |
| Anthropic | #6 |
| Wafer | #7 |
| xAI | #8 |
| TypeSafe AI | #9 |
| Google Vertex AI | #10 |
By token volume, among apps that opted in. Ranking order only.
| App | Rank |
|---|---|
| Cline | #1 |
| Command Code | #2 |
| Kilo Code | #3 |
| opencode | #4 |
| Hermes Agent | #5 |
| Zo Computer | #6 |
| Goblins | #7 |
| Ditto | #8 |
| Poke | #9 |
| Terac | #10 |
- Open-weight models handled 71.5% of gateway tokens in September but drew only about 23% of spend; their spend share has climbed from about 5% in June.
- DeepSeek's models carried 58% of tokens but 6.6% of spend; Moonshot's Kimi models were the reverse, at 3.6% of tokens and 13.2% of spend.
- Anthropic's share of gateway spend fell from 63.7% in August to 43.9% in September, the first time it dropped below half.
- Total gateway spend in September was about 13 times December 2025's level, while the average price per token fell 47%.
- 48% of teams changed their most-used model during September, an index record and the fifth straight monthly rise in switching.
- Jev, launched Sep 17, reached 35.3% of paid teams within two weeks, the highest reach in the index's history. The leaderboard's reach figure (22.3%) is lower because it counts all teams over a different window.
- 69.5% of spend on Chinese-origin models now runs through US-based providers, up from under 10% in December 2025.
Source: Vercel AI Gateway leaderboards for models, labs, providers and apps (vercel.com/ai-gateway/leaderboards, fetched 2026-11-04; per Vercel's documentation each share is for the most recent day in the trailing three-month window) and Vercel's "AI Gateway Production Index — October 2026" (vercel.com/blog/ai-gateway-production-index-october-2026). © 2026 Vercel. "AI Gateway Leaderboard Data". Licensed under CC BY 4.0.
Looking for AI risk / P(doom) data? That lives on its own board, since it mixes real incident counts and a prediction market rather than fitting the time-series/observatory shapes above — see the AI Risk Signal Board.
Coverage Leaderboard — top movers
Which tracked companies are getting the most (recency-weighted) mentions in machine.net’s own published coverage right now — see the methodology for how it’s computed.
| Company | Signal Score | 7d | Sector |
|---|---|---|---|
| Anthropic | 100 | +14.4 | Frontier Lab |
| Nvidia | 62 | +44.9 | Hardware |
| OpenAI | 58 | -41.7 | Frontier Lab |
| Meta | 53 | +0 | Social Platform |
| AMD | 47 | +0 | Hardware |