Pin official model source
Upstream repositories and LICENSE files can change.
Record full 40-character SHA, visible original checkpoint names and pinned card and config URLs.
Silent source drift or mixing different variants.
THINKING MACHINES LAB · FIELD VERIFIED
Primary sources last checked on 21 Nov 2026. Verification freshness is tracked separately from factual change, and unknown values remain explicit rather than inferred.
MODEL INTELLIGENCE · SOURCE-FIRST
Open Model Weights records Inkling 2 Mini as a 26.4B stored / 3.4B active per pass, publisher; mixed-precision 20.19GB official storage-parameter text and image input; text output open-weight model in the Inkling 2 family, with Thinking Machines Lab listed as the developer. The checked source repository provides 2 recognized weight artifacts, format evidence for Safetensors and a verified context window of 1,048,576 tokens, publisher-declared; no OMW runtime test. Values that are not evidenced in the checked sources remain explicitly undisclosed or unknown.
The repository declares Apache-2.0. Commercial use subject to Apache-2.0 conditions and any separately applicable publisher use policy. Open Model Weights reports the checked license evidence and classification; this is not legal advice.
No base model is declared in the checked standard lineage metadata, so Open Model Weights does not infer one. Training recipe status: not-independently-reproduced; data disclosure status: unknown.
+ Save to my shortlistMonitor changes →Local browser monitoring · no background email alerts
DIRECT ANSWER · FEATURED MODEL BRIEFING
Source-grounded editorial layer · reviewed 2026-11-21
Inkling 2 Mini is an original, publicly downloadable 26.4B stored / 3.4B active per pass, publisher; mixed-precision 20.19GB official storage release by Thinking Machines Lab. Its official Hugging Face repository is pinned at ed40e7da625f594fc9c4990f5f66be24873c22d7. The declared license is Apache-2.0 based on a publisher model-card declaration without a standalone LICENSE file. Open Model Weights has not executed inference or benchmarks for this checkpoint.
WHY IT MATTERS
Twelve middle decoder layers are reused three times: the 24 physical-layer model performs 48 layer applications, with 5.4B publisher-estimated effective activations including repeated passes.
The exact source-reported checkpoint contains 2 Safetensors files at the pinned revision, with 26.39 billion observed tensor values. This is a storage metadata count, not a runtime-memory or capability benchmark.
Repository declares Apache-2.0 through its official card but has NO standalone LICENSE in current API siblings; publisher also links a Model Acceptable Use Policy. Its 1M context is not a demonstrated 32GB-device workload.
MODEL-SELECTION VIEW
Inkling 2 Mini is an original, publicly downloadable 26.4B stored / 3.4B active per pass, publisher; mixed-precision 20.19GB official storage release by Thinking Machines Lab. Its official Hugging Face repository is pinned at ed40e7da625f594fc9c4990f5f66be24873c22d7. The declared license is Apache-2.0 based on a publisher model-card declaration without a standalone LICENSE file. Open Model Weights has not executed inference or benchmarks for this checkpoint.
Twelve middle decoder layers are reused three times: the 24 physical-layer model performs 48 layer applications, with 5.4B publisher-estimated effective activations including repeated passes.
Original checkpoint evidence is tied to https://huggingface.co/thinkingmachines/Inkling-2-Mini/tree/ed40e7da625f594fc9c4990f5f66be24873c22d7. The publisher's source inventory lists 2 Safetensors files, including model-00001-of-00001.safetensors and mtp.safetensors. This source review inventories file names and aggregate tensor metadata, not a full local checksum download or inference run.
Parameter labels and published precision must be separated from activation counts and load-time memory. 26.4B stored / 3.4B active per pass, publisher; mixed-precision 20.19GB official storage.
The precise license classification is Apache-2.0. No separate LICENSE file appears in the source inventory; the publisher model card declares Apache-2.0 and references an Acceptable Use Policy. Weight rights do not automatically license every dependency or dataset.
No independently verified base-model lineage was established by this source review.
Artificial Analysis independently scores Inkling 2 Mini at 20, but this is not a top-tier general-purpose frontier score and is not an OMW runtime test.
Repository declares Apache-2.0 through its official card but has NO standalone LICENSE in current API siblings; publisher also links a Model Acceptable Use Policy. Its 1M context is not a demonstrated 32GB-device workload.
Candidate workflows: Local coding, tool-oriented reasoning and looped-transformer research on suitably provisioned machines. Published runtime references mention Transformers, llama.cpp, MLX, SGLang. Running vendor-specific loaders may execute source code; inspect and pin dependencies.
Repeatable evaluation should pin the exact revision, model processor and chat/decision template; publish hardware, memory, latency, task outcomes and failure cases. No such Open Model Weights runtime measurements are claimed here.
IMPLEMENTATION & EVALUATION
Source-pinned field evidence; no OMW model inference or independently measured throughput.
Upstream repositories and LICENSE files can change.
Record full 40-character SHA, visible original checkpoint names and pinned card and config URLs.
Silent source drift or mixing different variants.
Downloadable weights do not automatically cover runtime code and commercial rights.
Inspect source-pinned license and any publisher usage policy, third-party notices or remote model code.
Licensing overreach or arbitrary code execution.
Storage and publisher benchmark numbers are not production capabilities.
Publish exact hardware, precision, runtime, batch settings, memory and latency for a held-out task.
Overclaimed context, memory or performance.
Leading requires an independently evaluated checkpoint-identical reference.
Link independent result to the actual source checkpoint, harness and evaluation date.
Unjustified Leading promotion based on API-only or publisher scores.
Proposed tests, not OMW benchmark results. Read the official model card at the pinned source revision ↗ Inspect OMW field evidence ↗
PRIMARY & INDEPENDENT SOURCES
FAQ
The official thinkingmachines/Inkling-2-Mini repository declares public, ungated Safetensors weights at immutable SHA ed40e7da625f594fc9c4990f5f66be24873c22d7; the checked manifest lists 2 original files.
No. This is an evidence-based source audit, not an OMW execution or independent performance ranking.
The recorded publisher license is Apache-2.0. Only the official model-card license declaration was found, with an additional policy reference. Check all conditions and separately licensed components.
The repository storage is approximately 20.19 decimal GB, not a measured GPU memory budget. Runtime operators, KV cache and media preprocessing add requirements.
Notable records differentiated primary-source releases; Leading requires matched, independent evidence for this exact checkpoint.
03 · PRACTICAL TECHNICAL PROFILE
Inspect weight-memory estimates, model lineage and runtime evidence separately from promotional or untested performance claims.
DEPLOYMENT PROFILE
These bars visualize weight-only decimal-GB estimates. They exclude KV cache, activations, runtime overhead, optimizer state and sharding.
LICENSE & USE
Commercial use subject to Apache-2.0 conditions and any separately applicable publisher use policy
This section summarizes the checked repository evidence. It is not legal advice. Inspect license evidence ↗
LINEAGE & PROVENANCE
Open Model Weights does not infer a hidden base model or training corpus when the checked sources do not declare one.
RUNTIME SIGNALS
Declared and model-card-mentioned runtimes are discovery signals. They are not treated as Open Model Weights runtime tests unless explicitly measured.
EVIDENCE TRAIL
This is where the article remains auditable: source links on one side, the registry's observed change history on the other.
Pinned ed40e7da625f594fc9c4990f5f66be24873c22d7; recorded official 2 weight-file names and original source license; independent performance not inferred.
04 · EVIDENCE AND UPDATES
Inspect dated source checks and any differences between published claims and the evidence; these checks are not independent model benchmarks.
SOURCE-BASED VISUAL BRIEF · NOTABLE
Three evidence-linked visual summaries: parameter scale, illustrative precision memory and verifiable provenance. No performance ranking is implied.
No reliable single-model weight-only memory figure.
FRESHNESS & VERIFICATION
A verification check does not pretend the model changed. Open Model Weights updates the verification date after a real primary-source check, while factual-update dates move only when retained source evidence changes.
CONFLICT INTELLIGENCE
05 · SOURCE RECORD
Full repository-level attributes, licensing, format details and provenance are available below without hiding or rewriting the original data.
01 · WEIGHTS
02 · LICENSE
Classification describes the checked license terms; it is not legal advice.
License / model-card evidence ↗03 · HARDWARE
Arithmetic for hypothetical whole-model uniform precision, not the exact published mixed NVFP4/MXFP8 artifact. Official repository storage: 20.19 GB including nonweight files; no runtime memory or speed measurements.
04 · FORMATS
“Not observed” means no matching repository artifact/name/dtype signal was found; it does not prove no third-party conversion exists.
05 · LINEAGE
06 · TRAINING ASSETS
07 · RUNTIME SUPPORT
Open Model Weights separates model-repository evidence from external runtime-release evidence. An external support statement does not change the official model repository's weight formats and does not imply an Open Model Weights inference benchmark.
08 · SOURCES
09 · VERIFICATION HISTORY
Pinned ed40e7da625f594fc9c4990f5f66be24873c22d7; recorded official 2 weight-file names and original source license; independent performance not inferred.
MODELS FOR AI AGENTS · ROLE EVIDENCE
Local assistant and tool model. A candidate for a reasoning or coordination role. No end-to-end agent performance is inferred from its model category.
The curated notable tier is not an agent-readiness rating. Model-specific tool-call success, long-horizon task accuracy and compatible runtime configurations are not independently established here.
Source-linked license: Apache-2.0 (allowed commercial-use classification; see original terms).
Context: 1048576 configured tokens.
BF16 weights: 52.78 estimated decimal GB, not runtime VRAM.
Compare agent model candidates → · Agent evaluation method → · Inspect original model repository ↗
These records are separately sourced. They do not replace independent model-plus-runtime tests.
No model-specific agent capability evaluation is recorded for this checkpoint. Missing evidence is unknown, not a failed task.
Read evidence details and benchmarks → · Use the agent model selector →