// live acquisition system
Model Radar
An automated sweep of Hugging Face and OpenRouter, every six hours — filtered to notable labs and high-signal releases, with a capability catalog mapping every contact to the jobs it can actually do.
// introductions
Last 30 days
- Aug 27, 2026inclusionai/inclusionAI: Ling 3.0 Flash Fin (free)
- Aug 26, 2026qwen/Qwen: Qwen3.8 Flash
- Aug 26, 2026z-ai/Z.ai: GLM 5.3 Flash
- Aug 21, 2026meta/Meta: Muse Spark 1.2 Contributor
- Aug 21, 2026deepseek/DeepSeek: DeepSeek V4 Flash Vision Exp
- Jul 16, 2026moonshotai/MoonshotAI: Kimi K3
- Jul 16, 2026meta/Meta: Muse Spark 1.1
- Jul 9, 2026openai/OpenAI: GPT-5.6 Luna Pro
- Jul 9, 2026openai/OpenAI: GPT-5.6 Luna Pro (batch)
- Jul 9, 2026openai/OpenAI: GPT-5.6 Luna
- Jul 9, 2026openai/OpenAI: GPT-5.6 Luna (batch)
- Jul 9, 2026openai/OpenAI: GPT-5.6 Terra Pro
- Jul 9, 2026openai/OpenAI: GPT-5.6 Terra Pro (batch)
- Jul 9, 2026openai/OpenAI: GPT-5.6 Terra
- Jul 9, 2026openai/OpenAI: GPT-5.6 Terra (batch)
- Jul 9, 2026openai/OpenAI: GPT-5.6 Sol Pro
$ +93 more contacts in the registry below — the long tail is mostly community fine-tunes.
// catalog
Latest model catalog by task type
Chat
chatGeneral-purpose assistant conversation and Q&A.
Vibecoding
codeCode generation, completion, and pair-programming.
Bots
botsHigh-volume conversational agents and support bots.
Claw
agentAgentic tool use — function calling, browsing, computer use.
Spark
fastSmall, fast, affordable tiers for everyday workloads.
Extraction
xtractStructured output, document parsing, retrieval work.
Prose
proseLong-form writing, editing, and style.
Vision
visImage understanding — reading, describing, reasoning over pixels.
Image
imgImage generation and editing.
Video
vidVideo generation and understanding.
Frontend
uiUI code, design-to-code, and layout work.
Gamedev
gameGame code, procedural content, and systems scripting.
Transactions
txnReal-world tasks — forms, commerce, multi-step execution.
scores: · serviceable · ● strong · ◆ elite — ed marks editorial curation; all other scores are heuristically inferred from model metadata, not benchmarks.
// matrix
Capability matrix
| model | chat | code | agent | fast | rsn | vis | lat | ctx |
|---|---|---|---|---|---|---|---|---|
| x-ai/SpaceXAI: Grok 4.5ed | ◆ | ● | ● | · | ● | ● | ● | ● |
| anthropic/Anthropic: Claude Fable 5.1ed | ◆ | ◆ | ◆ | · | ◆ | ● | · | ◆ |
| anthropic/Anthropic: Claude Fable 5.1 (batch)ed | ◆ | ◆ | ◆ | · | ◆ | ● | · | ◆ |
| Meta/Llama 4 Scouted | ● | ● | ● | ● | ● | ● | ● | ◆ |
| DeepSeek/DeepSeek V4ed | ● | ◆ | ● | · | ◆ | — | · | ● |
| Qwen/Qwen3.5 72B Instructed | ● | ◆ | ● | · | ● | — | · | ● |
| Mistral AI/Mistral Large 3ed | ◆ | ● | ● | · | ● | — | ● | ● |
| OpenAI/GPT-5.2ed | ◆ | ◆ | ◆ | · | ◆ | ◆ | · | ◆ |
| Google/Gemini 3 Flashed | ◆ | ● | ● | ◆ | ● | ◆ | ◆ | ◆ |
| anthropic/Anthropic: Claude Opus 4.8 (Fast)ed | ◆ | ◆ | ◆ | · | ◆ | ● | · | ◆ |
| anthropic/Anthropic: Claude Opus 4.8ed | ◆ | ◆ | ◆ | · | ◆ | ● | · | ◆ |
| anthropic/Anthropic: Claude Opus 4.8 (batch)ed | ◆ | ◆ | ◆ | · | ◆ | ● | · | ◆ |
| x-ai/SpaceXAI: Grok 4.6ed | ◆ | ● | ● | · | ● | ● | ● | ● |
| anthropic/Claude Opus 5 (Fast)ed | ◆ | ◆ | ◆ | · | ◆ | ● | · | ◆ |
// watchlist
Upcoming — next week
openai/GPT-5 Codex (next iteration)
OpenAI has been shipping coding-tuned refreshes on a tight cadence; watch the changelog and OpenRouter for a new gpt-5-codex variant.
source →google/Gemini 3 Pro
Flash shipped Aug 21; Pro historically follows within weeks. Expect image + long-context emphasis.
source →deepseek-ai/DeepSeek Flash tier
DeepSeek V4 landed Aug 22; a distilled fast/cheap tier would extend their price-pressure strategy. Watch HF org page.
source →anthropic/Claude Haiku refresh
Haiku-tier refreshes tend to follow flagship releases; no official date. Radar watching for new claude-* ids on OpenRouter.
source →Google DeepMind/Gemini-powered
Radar hint from Google DeepMind: "Empowering India’s next generation of innovators with ATL Saathi"
source →OpenAI Blog/GPT-6
Radar hint from OpenAI Blog: "GPT-6 Astra: The next generation in intelligence for work"
source →watchlist entries are signals, not commitments — "rumored" and "signal" confidence means no confirmed date exists.
// all contacts
Registry
inclusionai/inclusionAI: Ling 3.0 Flash Fin (free)
qwen/Qwen: Qwen3.8 Flash
z-ai/Z.ai: GLM 5.3 Flash
meta/Meta: Muse Spark 1.2 Contributor
deepseek/DeepSeek: DeepSeek V4 Flash Vision Exp
moonshotai/MoonshotAI: Kimi K3
meta/Meta: Muse Spark 1.1
openai/OpenAI: GPT-5.6 Luna Pro
openai/OpenAI: GPT-5.6 Luna Pro (batch)
openai/OpenAI: GPT-5.6 Luna
openai/OpenAI: GPT-5.6 Luna (batch)
openai/OpenAI: GPT-5.6 Terra Pro
openai/OpenAI: GPT-5.6 Terra Pro (batch)
openai/OpenAI: GPT-5.6 Terra
openai/OpenAI: GPT-5.6 Terra (batch)
openai/OpenAI: GPT-5.6 Sol Pro
openai/OpenAI: GPT-5.6 Sol Pro (batch)
openai/OpenAI: GPT-5.6 Sol
openai/OpenAI: GPT-5.6 Sol (batch)
x-ai/SpaceXAI: Grok 4.5
sakana/Sakana: Fugu Max
inclusionai/inclusionAI: Ling 3.0 Flash VL
inclusionai/inclusionAI: Ling 3.0 Flash VL (free)
deepseek/DeepSeek: DeepSeek V4.1 Flash
openai/OpenAI: GPT-6 Astra
openai/OpenAI: GPT-6 Astra (batch)
openai/OpenAI: GPT-6 Astra Pro
openai/OpenAI: GPT-6 Astra Pro (batch)
inclusionai/inclusionAI: Ling 3.0 Flash Sante (free)
qwen/Qwen: Qwen3.8 Max (0902)
meta/Meta: Muse Spark 1.3 Contributor
meta/Meta: Muse Spark 1.3
google/Google: Gemini 3.8 Flash
google/Google: Gemini 3.8 Flash (batch)
anthropic/Anthropic: Claude Fable 5.1
anthropic/Anthropic: Claude Fable 5.1 (batch)
ibm-granite/IBM: Granite 4.2 8B
tencent/Tencent: Hy4 preview
inclusionai/inclusionAI: Ling 3.0 Flash Fin
~z-ai/Z.ai: GLM Flash Latest
z-ai/Z.ai: GLM 5.3 Flash (batch)
deepseek/DeepSeek: DeepSeek V4 Flash Vision Exp (batch)
z-ai/Z.ai: GLM 5.3 (batch)
qwen/Qwen: Qwen3.8 2.4T A95B (batch)
deepseek/DeepSeek: DeepSeek V4 Pro 0813 (batch)
meta/Meta: Muse Glimmer 30B (batch)
deepseek/DeepSeek: DeepSeek V4 Flash 0731 (batch)
thinkingmachines/Thinking Machines: Inkling Small (batch)
moonshotai/MoonshotAI: Kimi K3 (batch)
~openai/OpenAI: GPT Astra Latest
~openai/OpenAI: GPT Sol Latest
~openai/OpenAI: GPT Terra Latest
~openai/OpenAI: GPT Luna Latest
sakana/Sakana: Fugu Ultra v2
Meta/Llama 4 Scout
Mixture-of-experts open-weights flagship with a 10M-token context window, tuned for agentic workloads.
DeepSeek/DeepSeek V4
Reasoning-first MoE with strong code and math benchmarks at a fraction of closed-model inference cost.
Qwen/Qwen3.5 72B Instruct
Dense 72B instruct model with best-in-class multilingual tool use; available on HF and OpenRouter.
Mistral AI/Mistral Large 3
Multimodal flagship targeting enterprise agents with native document and image understanding.
OpenAI/GPT-5.2
OpenAI's latest frontier model, routed via OpenRouter with a 400K context window.
Google/Gemini 3 Flash
Low-latency multimodal model with 1M context, positioned for high-throughput agent fleets.
qwen/Qwen: Qwen3.7 Plus
minimax/MiniMax: MiniMax M3
minimax/MiniMax: MiniMax M3 (batch)
minimax/MiniMax: MiniMax M3 (free)
stepfun/StepFun: Step 3.7 Flash
anthropic/Anthropic: Claude Opus 4.8 (Fast)
anthropic/Anthropic: Claude Opus 4.8
anthropic/Anthropic: Claude Opus 4.8 (batch)
qwen/Qwen: Qwen3.7 Max
nvidia/NVIDIA: Nemotron 3.5 Content Safety (free)
nvidia/NVIDIA: Nemotron 3 Ultra
nvidia/NVIDIA: Nemotron 3 Ultra (batch)
nvidia/NVIDIA: Nemotron 3 Ultra (free)
tencent/Tencent: Hy-MT2-1.8B
tencent/Tencent: Hy-MT2-30B-A3B
~z-ai/Z.ai: GLM Latest
tencent/Tencent: Hy-MT2-7B
z-ai/Z.ai: GLM 5.3
qwen/Qwen: Qwen3.8 27B
google/Google: Gemini 3.7 Flash
google/Google: Gemini 3.7 Flash (batch)
bytedance-seed/ByteDance Seed: Seed 2.1 Turbo
qwen/Qwen: Qwen3.8 2.4T A95B
bytedance-seed/ByteDance Seed: Seed-2.0-Code
deepseek/DeepSeek: DeepSeek V4 Pro 0813
x-ai/SpaceXAI: Grok 4.6
nvidia/NVIDIA: Nemotron 3.5 Lightning
nvidia/NVIDIA: Nemotron 3.5 Lightning (free)
sakana/Sakana: Sakana Namazu
upstage/Upstage: Solar Pro 4
meta/Meta: Muse Glimmer 30B
meta/Meta: Muse Spark 1.2
qwen/Qwen: Qwen3.8 Max
~deepseek/DeepSeek V4 Flash Latest
deepseek/DeepSeek: DeepSeek V4 Flash 0731
thinkingmachines/Thinking Machines: Inkling Small
thinkingmachines/Thinking Machines: Inkling Small (free)
qwen/Qwen: Qwen3.7 Flash
anthropic/Claude Opus 5 (Fast)
anthropic/Claude Opus 5
anthropic/Claude Opus 5 (batch)
inclusionai/Ling-3.0-flash
google/Google: Gemini 3.6 Flash
google/Google: Gemini 3.6 Flash (batch)
google/Google: Gemini 3.5 Flash Lite
google/Google: Gemini 3.5 Flash Lite (batch)
thinkingmachines/Thinking Machines: Inkling
thinkingmachines/Thinking Machines: Inkling (batch)
thinkingmachines/Thinking Machines: Inkling (free)
$ how the radar works
A scheduled job queries the Hugging Face model API (sorted by creation date) and the OpenRouter model list, then keeps only notable contacts: models from known lab orgs, or community releases that clear an engagement bar and aren't derivative repacks (see content/models/notable-orgs.json). Each contact gets a detail page and a heuristic capability profile; editors can override any score in content/models/capabilities.json. The watchlist combines curated entries with radar hints extracted from the news sweep.