Inkling-Small

Released
2026-07-30
Last refreshed
2026-10-05
Status
Researched 1d ago
Open sourceCommercial use: permittedMultimodalAgentsVisionJSON / Tool use

Inkling-Small is available now for agents, vision, and json / tool use with open-source; evaluate it while provider pricing coverage matures.

Thinking Machines Lab releases · 2 in the last 12 monthsChangelog →
Specifications
Family
Inkling
Released
2026-07-30
Parameters
276B (12B activated)
Architecture
Mixture of Experts
Specialization
general
Openness
Open source
License
Apache 2.0OSI-approvedCommercial use: permitted
Weights
Available
Code
Unknown
Training
Multi-stage
Created by

Building collaborative AI aligned with human values.

San Francisco, California, United States
Founded 2025
Website
Pricing

No tracked provider token pricing is available yet.

About

Thinking Machines Inkling-Small (HF thinkingmachines/Inkling-Small) is the smaller open-weight sibling of Inkling, announced 30 July 2026. It is a general-purpose multimodal model that takes text, image (40-4096px) and 16 kHz WAV audio input and produces text. Architecture: 42-layer decoder-only sparse MoE with 276B total and 12B active parameters (6 of 256 experts plus 2 shared) and hybrid local/global attention, shipped in BF16 and NVFP4. Apache 2.0. Hosted by DeepInfra, Together and Baseten per HF provider mapping.

Provider price ladder

No tracked provider token pricing is available for this model yet.

Capabilities

VisionMultimodalReasoningJSON / Tool useAudio

Benchmark peer barsfor Agents

No task-mapped benchmark peers are available for this model yet.

Benchmark scores(3)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.
BenchmarkScoreVersionEvaluationSource
Google-Proof Q&A88.5diamondObserved 2026-08-14
Configuration: Inkling-Small (Extra High)
Harness: Epoch AI Inspect harness (GPQA Diamond, 198 Q)
Evaluator: Epoch AI
Confidence: confirmed
Notes: Epoch AI-run; benchmark_data.zip gpqa_diamond.csv row 'inkling-small_xhigh' 0.88510 (run 2026-08-14). CC BY 4.0, credit Epoch AI.
Source
FrontierMath Tier 417.1FrontierMath Tier 4 v2 (private)Observed 2026-08-14
Configuration: Inkling-Small (Extra High)
Harness: Epoch AI FrontierMath Tier 4 v2 private set
Evaluator: Epoch AI
Confidence: confirmed
Notes: Epoch AI-run; frontiermath_tier_4_v2.csv row 'inkling-small_xhigh' 0.17073 (run 2026-08-14). CC BY 4.0.
Source
ARC-AGI-240.1ARC-AGI-2 semi-privateObserved 2026-10-01
Configuration: Inkling-Small (XHigh)
Harness: ARC Prize semi-private evaluation
Evaluator: ARC Prize Foundation
Confidence: confirmed
Notes: ARC Prize official data https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01) modelDisplayName 'Inkling Small (XHigh)' score 0.40139; matches Epoch AI mirror. Highest published effort.
Source

Migration checks

No linked migration route is available for this model yet.

API versions

thinkingmachines/Inkling-SmallInkling Small

Rankings & picks(1)