Release date
When the model appeared, what shipped at launch, and what is still outstanding. The top of this page is rewritten automatically from the vendor's own sources.
Inkling 2 Small was released on Nov 5, 2026 by Thinking Machines Lab. It launched as a closed-weights model available through Tinker for inference and fine-tuning, with the company stating that full Apache 2.0 weights would follow about a week later — targeted for around Nov 12, 2026.
Current state
API-only (weights pending)- Released
- Nov 5, 2026 · 2 days ago
- Weights available
- No
- Weights expected
- Nov 12, 2026 · in 5 days
- Checkpoints downloadable
- Not yet
- Last checked
- Nov 6, 2026, 11:44 UTC
What the state is based on
| Signal | Observed | Detail |
|---|---|---|
tinker-listed | Yes | Inkling 2 Small is listed in the Tinker model catalogue, so API inference and fine-tuning are live. |
license-closed | Yes | Model card lists the license as "Tinker Terms of Use". |
apache-intent | Yes | We intend to make the full weights available under an Apache 2.0 license next week after the completion of external safety testing and red-teaming. |
hf-org-reachable | Yes | Hugging Face org listing returned 6 models. |
hf-repo-present | No | No Inkling 2 Small weights repository in the thinkingmachines organisation yet. |
expected-date-passed | No | Announcement points at roughly Nov 12, 2026 for the weights release (6 days out). |
Timeline
Inkling 2 Small is the second generation of Thinking Machines' model family. The first generation shipped in two sizes earlier in 2026: Inkling — 975B parameters, 41B active — in mid-July, and Inkling-Small — 276B parameters, 12B active — at the end of July. Both were open-weights releases.
The gap between launched and released
There are two dates that matter and they are easy to conflate. The model launched on Nov 5, 2026: the API, the playground, the model card and the benchmark table all went live that day. The weights are a separate event, deliberately sequenced after external safety testing.
Until the weights arrive, "can I run it locally" has a single answer: no. Inkling 2 Small is only reachable through Tinker. Once they do arrive, the constraint shifts from access to hardware — see memory requirements, where even a 4-bit quantisation needs roughly 380.7 GB.
Independent measurement at launch put the model at 135 output tokens per second and 18.5 minutes per Intelligence Index task — slow relative to its peers, which is worth knowing before you plan a latency-sensitive product around it.