AI Model Radar / AI模型雷达

Key Events

Key Events are industry changes supported by multiple public sources, not an audit system.

Key Event

OpenRouter data confirms 67 available catalog versions

The bounded OpenRouter public-model data maps 67 catalog versions as available, with 90 remaining not found and zero needs-review rows.

officialOpenRouter1 sources
Involved

OpenAI / Anthropic / Google / Meta / DeepSeek / Alibaba / Qwen

Why it matters

Availability and pricing can be shown as a real lane while unmatched rows stay visible as source gaps.

Evidence caveat: Not-found means absent from this public data, not proof that a provider cannot serve the model elsewhere.

Key Event

Qwen image and Qwen3 model cards add high-volume public activity evidence

Hugging Face counters for Qwen-Image and Qwen3 family repositories were consolidated into adoption and Public Signals rows.

Hugging Faceofficial organization3 sources
Involved

Alibaba / Qwen · Qwen Image / Qwen3 235B A22B / Qwen3 VL 235B A22B

Why it matters

The event links model-family activity to concrete source counters without turning them into global adoption claims.

Evidence caveat: Hugging Face counters are platform-local and can reflect curiosity as well as production use.

Key Event

DeepSeek open-weight family evidence is consolidated as ordinary public activity

DeepSeek V3.2, R1, V4, Janus Pro, and VL2 source-backed rows were consolidated from public model cards and official parameter disclosures.

Hugging Faceofficial organization4 sources
Involved

DeepSeek · DeepSeek V3.2 / DeepSeek R1 / Janus Pro / DeepSeek VL2

Why it matters

It keeps DeepSeek comparable with Qwen, GLM, Kimi, Hunyuan, and other tracked companies under the same source rules.

Evidence caveat: Public model-card activity is not a global usage total.

Key Event

Qwen3-TTS parameter disclosure enters reviewable model evidence

The Qwen3-TTS README discloses a 1.7B model scale for the series; API flash rows are filled with a SKU caveat.

official GitHubmanual review1 sources
Involved

Alibaba / Qwen · Qwen3 TTS

Why it matters

Parameter coverage improves while preserving a review path for exact API SKU mapping.

Evidence caveat: Official disclosure is accepted for the series, but the API SKU mapping remains medium-confidence.

Key Event

Gemma and EmbeddingGemma repositories add public open-model activity signals

Google open-model repositories including Gemma 3n, EmbeddingGemma, and FunctionGemma are represented with bounded Hugging Face counters.

Hugging Faceofficial organization3 sources
Involved

Google · Gemma 3n E4B IT / EmbeddingGemma 300M / FunctionGemma 270M IT

Why it matters

Open-model adoption and Public Signals can be compared against other first-class tracked companies with visible caveats.

Evidence caveat: Counters are platform-specific and should not be read as total usage.

Key Event

Meta vision tooling evidence combines SAM 3 parameters with DINOv3 repository activity

SAM 3 parameter disclosure and DINOv3 GitHub public activity are represented as one key vision-tooling event.

Hugging FaceGitHub2 sources
Involved

Meta · Segment Anything Model 3 / DINOv3

Why it matters

Event-level multi-source merge keeps related open-model and tooling signals together instead of scattering them across rows.

Evidence caveat: Repository stars and forks are public attention indicators, not deployment proof.