Overview
Inkling 2 Small is the first model of Thinking Machines Lab's second Inkling generation, announced on November 5, 2026. It is a sparse Mixture-of-Experts transformer with 594 billion total and 17 billion active parameters. It accepts text and image input, supports variable reasoning effort, and does not support audio.
At launch the model is available through Tinker, Thinking Machines' platform, with closed weights. The lab says it intends to publish the full weights under Apache 2.0 the following week, after external safety testing and red-teaming. That is a stated intention, so the license and open-weights flag on this page change only when the weights are out.
The model card states a context window of up to 1M tokens. Training used NVFP4 pretraining on NVIDIA GB300 NVL72 racks and a nine-day reinforcement-learning run on 2,048 GPUs. The announcement reports a score of 31 on the Artificial Analysis Intelligence Index.
| Released | 2026-11-05 |
|---|---|
| License | Tinker Terms of Use (Apache-2.0 weights planned) |
| Weights | API only |
| Parameters | 594B total · 17B active |
| Context | 1M |
| Architecture | Sparse Mixture-of-Experts transformer with a 292M-parameter vision encoder based on Perception Encoder |
| Modalities | Text, Vision |
| Status | Announced |
Strengths
- Up to 1M tokens of context with compaction for long agent trajectories
- Text and image input with variable reasoning effort
- Reported coding results include 72.1% on DeepSWE v1.1 and 86.3% on SWE-Bench Pro v2 (Hard)
- Internal misbehavior rate of 14.9, against 66.8 for the first Inkling (lower is better)
Best for
- Reach for it to try a new US lab model in the Tinker Playground or to fine-tune and sample it on Tinker.
- Reach for it for long-horizon coding agents that need a very long context.
- Wait for the Apache 2.0 weights if you need to self-host or fine-tune outside Tinker.
How to access
| Provider | Model ID |
|---|---|
| Tinker (Thinking Machines) ↗ | — |
Inkling — every version
The full lineage of the Inkling line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.
| Version | Released | Context | License |
|---|---|---|---|
| Inkling 2 Small | 2026-11-05 | 1M | Tinker Terms of Use (Apache-2.0 weights planned) |
| Inkling-Smallcurrent | 2026-07-30 | 1M | Apache-2.0 |
| Inkling | 2026-07-15 | 1M | Apache-2.0 |
FAQ
What is Inkling 2 Small?
Inkling 2 Small is Thinking Machines Lab's second-generation model, announced on November 5, 2026. It is a 594B total / 17B active sparse Mixture-of-Experts model with text and image input and a context window of up to 1M tokens, available on Tinker.
Are the Inkling 2 Small weights open?
Not at launch. Inkling 2 Small starts on Tinker with closed weights. Thinking Machines Lab says it intends to release the full weights under Apache 2.0 the following week, after external safety testing and red-teaming. No exact date is confirmed.
Does Inkling 2 Small support audio?
No. Inkling 2 Small accepts text and image input and generates text. The earlier Inkling-Small also took audio input, so audio is a difference between the two models.