✳ the wire · intel

New on the radar: Inkling 2 Mini appeared in OpenRouter catalog

Inkling 2 Miniconfirmedby ArtificialWatch

An email when every new Thinking Machines model answers on a public API, about 15 minutes after members, plus every frontier launch · no card

CONFIRMED2026-11-21 · Inkling 2 MiniInkling 2 Mininew on the radarINTEL
verified against a primary source

Our hourly sweep found an id that was not there an hour ago.

Catalog id: thinkingmachines/inkling-2-mini
Listed price: $0.10/$0.40 per 1M
Context window: 1,048,576 tokens
First observed by our sweep: 2026-11-21 03:42 UTC
Source: OpenRouter catalog

This entry is machine-written from what the sweep actually read, and nothing else. A catalog listing is a primary source for exactly one narrow fact — that this id now exists, at this price, with this context window — and it is not evidence of anything beyond that. It does not tell you the model is good, that the listing is intentional, that the weights are open, or that the vendor has announced it. Those questions need a human and an editor's judgement, and this item deliberately does not answer them.

It maps to Inkling 2 Mini, already on our watchlist — registry note: live — Nov 20 · SEARCH ONLY, NOT READ BY OUR SWEEP: thinkingmachines.ai, huggingface.co and every vendor domain are blocked from the sweep container and WebFetch is dead there, so the launch post was never opened; our own seenIds carries `inkling`, `inkling2small`, `inkling2smallfree` and `inklingsmall` and NO inkling-2-mini id in any catalog we read, so no alert fired and none was fired by hand · what the results agree on: a looped transformer that runs its middle layers three times per token, open weights on Hugging Face under Apache 2.0 (the variable-looped variant too), also on Tinker for inference and fine-tuning, and runnable locally — any Mac with 32GB+ RAM via Ollama, and NVIDIA DGX Spark per one source · Thinking Machines frames it as the answer to users who found Inkling 2 Small, at 594B, too big · PARAMETER COUNT IS UNSETTLED and no single figure should be quoted: the lab is reported at 26B total / 5.4B active with the 5.4B counting all three passes, while other outlets give 26.4B total / 3.4B active and one says 3.5B active · Artificial Analysis places it at 20 on its Intelligence Index, against 31 for Inkling 2 Small · the lab claims a lead on several agentic coding evaluations among open-weight models under 100B total, and says it is NOT the strongest small model for multi-turn chat or many-constraint instruction following — self-reported, unverified from here.

When a sighting like this turns out to be a launch, members hear it first.

The radar reads the public model catalogs every hour and the provider APIs every 60 seconds. Signal texts you the minute a launch is confirmed; free alerts follow by email about 15 minutes later.

Try Signal 7 days free →

sweeping every 60 seconds

Hear when every new Thinking Machines model drops — not when we write it up.

Claude Opus 5 went live at 16:51 UTC. Members had the alert at 16:52. Free gets a Chrome push and an email about 15 minutes after members.

  • Free forever
  • No card
  • Unsubscribe in one click

Want it the minute it answers? Try Signal 7 days free →

← back to the wire · every story, by month