Inference
EN
Thinking Machines Lab

Inkling-Small

Inkling-Small is a reasoning model from Thinking Machines Lab in the Inkling family. 35 benchmarks count toward its index, in 7 categories.

  • Open weights
  • Reasoning
  • Input: text · image · audio
  • Released 2026-07
RankedEnough results for an index.
IndexAll categories, out of 100.55.1±3.2Median 51.0
CoverageIndex weight that has results.95%
SpeedOutput tokens per second.80/sMedian 67/s
First tokenSeconds to the first output token.0.7 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.45Median $0.60
Output / 1MUS dollars per 1M output tokens.$1.20Median $2.55
ContextMaximum tokens in one request.524KMedian 262K
EloLMArena rating and rank.1414±5 #137Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
thinkingmachines/inkling-small
Max outputMaximum output tokens per request.
262K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
No
Cache read / 1MUS dollars per 1M cached tokens.
$0.10
WeightsLink to the published weights.
Hugging Face (opens in a new tab) ↗
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
3 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
Deep InfraHost$0.45$1.20524K
OpenRouterRouter$0.45$1.20524K
Arcee AIDeprecatedHost$0.50$1.20262K

US dollars per 1M tokens, list price. From models.dev (opens in a new tab) (MIT).

Run locally

265.96B parameters
16-bit
559.5 GB
8-bit
297.7 GB
4-bit
170.3 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy needs more than 128 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.7 counted55.7100 / 218
CodingCode writing and repair.8 counted54.9100 / 209
ReasoningLogic problems and puzzles.5 counted55.781 / 227
MultimodalTasks with images and text.3 counted54.756 / 84
KnowledgeFacts and expert knowledge.5 counted52.0120 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted65.8—
MathMath problems.6 counted55.363 / 168

Results

35 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Instruction Following BenchmarkInstruction82.2%4 / 57
Software Engineering Benchmark VerifiedCoding80.2%19 / 114
LiveCodeBenchCoding85.9% ±1.025 / 142
Artificial Analysis GPQA DiamondKnowledge89.5%43 / 199
SWE-benchCoding82.2% ±1.720 / 87
Graduate-Level Google-Proof Q&AKnowledge89.5%33 / 141
GPQA diamond · Epoch AIKnowledge88.5% ±1.953 / 217
Scientific Code BenchmarkCoding48.7%9 / 35
OTIS Mock AIME 2024-2025Math90.0% ±4.062 / 200
Humanity's Last ExamKnowledgeAbove fitFar above this model's index.47.8%24 / 77
Artificial Analysis Humanity's Last ExamKnowledge33.3%70 / 222
MCP AtlasAgentic79.6%17 / 52
Agent Arena command recoveryAgentic5.417 / 50
Critical Physics TasksReasoning8.3%74 / 214
AIME 2026Math95.5%12 / 34
Harvard-MIT Mathematics Tournament February 2026Math90.2%10 / 27
GPQA Diamond · BenchLMKnowledge89.5%33 / 84
ARC-AGI-2 (semi-private)Reasoning40.1%53 / 131
MMLU ProKnowledge85.6% ±0.355 / 135
GPQA Diamond · Vals AIKnowledge83.6% ±2.055 / 135
ARC-AGI-1 (semi-private)Reasoning84.0%52 / 126
Artificial Analysis Long Context ReasoningReasoning75.7%92 / 217
GDPval-AA normalizedAgentic31.2%60 / 133
Artificial Analysis Omniscience AccuracyKnowledge31.8%74 / 164
Artificial Analysis SciCodeCoding49.7%53 / 114
BrowseCompAgentic77.4%33 / 63
Artificial Analysis MMMU-ProMultimodal74.0%61 / 113
CharXiv ReasoningMultimodal81.3%29 / 49
SWE-bench ProCoding55.9%52 / 83
FrontierMath-Tiers-1-3-v2-PrivateMath46.3% ±3.057 / 90
Terminal-Bench 2.1 · Vals AIAgentic55.1% ±2.349 / 75
Humanity's Last Exam without toolsKnowledge31.6%36 / 53
Terminal-Bench 2.1 · BenchLMAgentic64.7%59 / 86
FrontierMath-Tier-4-v2-PrivateMath17.1% ±5.952 / 72
Massive Multi-discipline Multimodal Understanding ProMultimodal74.0%39 / 53
Code MigrationCoding13.7% ±3.859 / 77
Vibe Code Bench v1.1Coding19.1% ±3.495 / 123
SimpleQA VerifiedKnowledge19.1% ±1.279 / 91
Toolathlon-VerifiedAgentic54.4%20 / 23
CharXiv Reasoning without toolsMultimodal77.4%15 / 17
SkillsBenchCoding33.6% ±3.732 / 35
Mystery Game PuzzlesReasoning6.0% ±2.474 / 80
ProofBench v1.1Math6.0% ±2.451 / 54
Terminal-Bench 4.0Agentic0.0%59 / 61
IOICoding9.3% ±4.544 / 45
Agent Arena task outcomeAgentic−27.653 / 54
Agent Arena steerabilityAgentic−27.754 / 54
CorpFin v2Finance69.6% ±0.96 / 131
TaxEval v2Finance75.5% ±0.818 / 143
ProgramBenchCoding0.5% ±0.519 / 65
MedScribeHealthcare84.1% ±1.938 / 109
ARC-AGI-2 (public eval)Reasoning39.3%46 / 122
ARC-AGI-1 (public eval)Reasoning89.4%47 / 116
LegalBenchLegal83.0% ±0.562 / 149
MortgageTaxFinance62.3% ±1.055 / 97
Public Benefits Bench v1.1Public benefits59.4% ±1.334 / 50
MedCodeHealthcare37.9% ±2.274 / 106
Legal Research BenchLegal25.5% ±3.063 / 90
CyberBench v1.1Other64.1% ±6.235 / 48
Tax Agent BenchFinance15.4% ±2.460 / 82
Finance Agent (v2)Finance41.3% ±0.862 / 79
EMBFinance31.8% ±3.064 / 75
HLAB-Vals-1.1Legal2.9% ±1.434 / 35
Artificial Analysis EnterpriseOps-GymAgentic42.7%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseEpoch AICC BYVals AINo licence statedARC Prize FoundationTerms restrict reuseLMArenaCC BY 4.0models.devMIT

History

5 releases
Index56.1 → 55.1
23 Oct 202612 Nov 2026
Output / 1M$1.20 → $1.20
23 Oct 202612 Nov 2026

Questions

How much does Inkling-Small cost?
$0.45 per 1M input tokens, $1.20 per 1M output tokens.
How good is Inkling-Small?
Its index is 55.1 out of 100. That is place 102 of 283 ranked models.
What is Inkling-Small best at?
Reasoning: 55.7 out of 100, place 81 of 227.
How fast is Inkling-Small?
80 output tokens per second, 0.7 s to the first token.
What is the context window of Inkling-Small?
524K tokens.
Is Inkling-Small open source?
Yes. Its weights are published.
When was Inkling-Small released?
July 2026.

Badge

Inkling-Small on Inference360: place 102 of 283, index 55.1
HTML
<a href="https://inference360.ai/models/inkling-small"><img src="https://inference360.ai/badges/inkling-small.svg" alt="Inkling-Small on Inference360: place 102 of 283, index 55.1" height="20"></a>
Markdown
[![Inkling-Small on Inference360: place 102 of 283, index 55.1](https://inference360.ai/badges/inkling-small.svg)](https://inference360.ai/models/inkling-small)

Compare with

Step 3.7 Flash55.3GPT-555.0GPT-5.155.3Gemini 3 Flash55.0Inkling54.8

Lower price

dots3-note Preview63.4 · FreeApodex 1.1 Mini55.7 · FreeGLM-5.3-Flash63.1 · $0.078/1M

More from Thinking Machines Lab

All models
Inkling54.8Inkling 2 Small66.4