Passer au contenu principal
AITier

Inkling Small

Thinking MachinesOpen WeightApache 2.0 · Usage Commercial

Description

Inkling-Small is Thinking Machines Lab's efficient open-weights MoE multimodal model (276B total / 12B active parameters) released under Apache 2.0. It accepts text, image, and audio inputs and generates text, with native reasoning, variable thinking effort, and a context window up to 1M tokens (Tinker exposes 64K and 256K configurations). Hugging Face weights: thinkingmachines/Inkling-Small and thinkingmachines/Inkling-Small-NVFP4. Vendor self-host VRAM: BF16 ≥ ~600 GB aggregated; NVFP4 ≥ ~180 GB aggregated. SWE-Bench Verified 80.2% vs Inkling 77.6% (same bash-only harness). Fine-tuning and playground chat are available via Tinker. Official Tinker serverless inference (256K, list): $0.30 / $1.20 per 1M input/output tokens ($0.06 cached).

Date de sortie
2026-07-30
Paramètres
276.0B
Longueur du contexte
1.0M
Modalités
audio, image, text

Radar de capacités

33
general
52
coding
90
reasoning
64
science
50
agents
80
multimodal

Classements

Domaine#RangScoreSource
Capacité agentique28
53.0
LS
Classement codage121
72.0
AA
Classement général191
56.0
AA
Classement multimodal67
49.0
LS
Science89
73.0
AA

Scores de benchmarks (LLM Stats)

(LLM Stats (zeroeval))

Agents

AA-Briefcase917.00 / 3000Aut.
Toolathlon54.4%Aut.
Tau3 Banking15.5%Aut.

Audio

MMAU77.0%Aut.

Chat

VoiceBench Avg90.1%Aut.

Factuality

SimpleQA Verified20.6%Aut.

General

GDPval-AA1269.00 / 3000Aut.
Global-MMLU-Lite86.7%Aut.
Artificial Analysis40.0%Aut.

Instruction Following

IFBench82.2%Aut.

Math

AIME 202695.5%Aut.

Reasoning

GPQANYU + Cohere + Anthropic (2023)89.5%Aut.
ARC-AGI84.0%Aut.
SWE-Bench Verified80.2%Aut.
MCP Atlas79.6%Aut.
CharXiv-R77.4%Aut.
BrowseCompOpenAI (2025)77.4%Aut.
Terminal-Bench 2.164.7%Aut.
SWE-Bench ProPrinceton NLP (2024)55.9%Aut.
SciCode48.7%Aut.
ARC-AGI v240.1%Aut.
Humanity's Last Exam31.6%Aut.
CritPT8.3%Aut.

Vision

MMMU-Pro74.0%Aut.

Indices d'évaluation AA

(Artificial Analysis)
Gpqa(NYU + Cohere + Anthropic (2023))
89.5
Lcr(Artificial Analysis)
75.7
Terminalbench V2 1
55.1
Coding Index(Artificial Analysis)
52.9
Scicode(UIUC + Argonne National Lab (2024))
49.7
Hle(Center for AI Safety + Scale AI (2025))
33.3
Intelligence Index(Artificial Analysis)
32.3
Tau Banking
18.8

Scores par catégorie LLM Stats

(LLM Stats (zeroeval))
Legal
100
Finance
100
Productivity
100
Agents
100
Reasoning
100
General
100
Language
90
Multimodal
80
Search
80
Instruction Following
80
Frontend Development
80
Audio
80
Physics
70
Biology
70
Chemistry
70
Code
70
Spatial Reasoning
60
Vision
60
Math
50
Tool Calling
50

Tarification

Prix d'entrée$0.3 / 1M tokens
Prix de sortie$1.2 / 1M tokens
Prix mixte (3:1)$0.525 / 1M tokens
Prix de lecture cache$0.1 / 1M tokens

Vitesse

Tokens/sec117.2
Délai du premier token2.19s
Temps de réponse19.25s

Classement des Prix par Fournisseur

Classement des Prix par Fournisseur

8 fournisseurs

Moins cher: Thinking Machines LabPlus cher: LLMTR
FournisseurEntréeSortie
1Thinking Machines LabMoins cher
$0
$0
2Thinking MachinesPRINCIPAL
$0.3
$1.2
3OpenRouter
$0.45
$1.2
4Kilo Gateway
$0.45
$1.2
5Baseten
$0.5
$1.2
6Vercel AI Gateway
$0.5
$1.2
7Arcee
$0.5
$1.2
8LLMTR
$0.58
$1.44

Comparer les prix entre différents fournisseurs API pour ce modèle.

Sources externes

LLM StatsArtificial Analysis