Thinking Machines: Inkling Small

by Thinkingmachinestext+image+audio->text3 endpoints
Neura Intelligence Index
63.2/ 100
Rank
#119
Confidence
high
Benchmarks
11(28% coverage)

Overview

Inkling Small is an open-weight multimodal mixture-of-experts model from Thinking Machines Lab, with 12B active parameters out of 276B total. It is positioned as the smaller, more efficient member of the Inkling family and is suited for reasoning, coding, agentic workflows, retrieval-augmented generation, instruction following, and multilingual conversation.

Capabilities

Text generation
Image understanding
Audio processing
Tool use / Function calling
Extended reasoning
Prompt caching
Long context
Top-P sampling
Stop sequences
Deterministic seed
Log probabilities

Modalities

Input
TextImageAudio
Output
Text

Technical Specifications

Context Window
1.0M tokens
Max Output
262.1K tokens
Tokenizer
Other
Uptime (24h)
9974.7%
Parameters
276000000000B
License
apache_2_0Open Weights
Country
🇺🇸US
Release Date
July 30, 20261mo

Supported Parameters

frequency_penaltyinclude_reasoninglogit_biaslogprobsmax_tokensmin_ppresence_penaltyreasoningreasoning_effortrepetition_penaltyseedstoptemperaturetool_choicetoolstop_ktop_logprobstop_p

Benchmark Performance

coding

BenchmarkScoreSource
SWE-bench Verified
80.2
verified
SWE-Bench Pro
55.9
verified
SciCode
48.7
verified

reasoning

BenchmarkScoreSource
GPQA Diamond
89.5
verified
ARC-AGI v2
40.1
verified
Humanity's Last Exam
31.6
verified

multimodal

BenchmarkScoreSource
CharXiv Reasoning
77.4
verified
MMMU-Pro
74
verified

agent

BenchmarkScoreSource
MCP Atlas
79.6
verified
BrowseComp
77.4
verified
Toolathlon
54.4
verified

Performance Over Time

Pricing

Per 1M tokens
Input$0.45
Output$1.20
Cache read$0.10
Added
July 30, 2026
Last synced 9/21/2026