Inkling-Small

Inkling-Small places in the top three on 2 benchmarks.

Thinking Machines Released July 15th, 2026 Compare with others

Benchmarks with a score

Ranked first, out of five or more models

In the top three

Thinking Machines Jul 2026

Inkling-Small

Open weights

150.8 #67 of 256

Epoch Capabilities Index

Self-Reported Index 149.8 #97 of 404

MMMU-Pro 74% #43 /90
IFBench 82.2% #3 /80
SciCode 48.7% #12 /49

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

Benchmark	Score	Rank	Source
IFBench	82.2%	#3 of 80	Lab model card
Humanity's Last Exam	47.8%	#50 of 154	Lab model card
LMArena (Elo)	1,405	#125 of 335	Official leaderboard
SEA-HELM	71.5	#29 of 71	Official leaderboard
StudyArena Text	1,227	#48 of 85	Official leaderboard
AA-Omniscience	-9.0	#7 of 10	Lab model card
Creative Writing v3	58.9	#85 of 120	Official leaderboard
Global MMLU Lite 3.0	86.7%	#3 of 4	Lab model card
SimpleQA Verified	19.1%	#83 of 94	Epoch AI run
OpenAI MRCR v2	10%	#44 of 47	Official leaderboard

Math and science

Benchmark	Score	Rank	Source
AIME 2026	95.5%	#9 of 38	Lab model card
GPQA Diamond	88.5%	#59 of 229	Epoch AI run
OTIS Mock AIME 2024-2025	90%	#69 of 212	Epoch AI run
ARC-AGI-2	40.1%	#46 of 98	Official leaderboard
ARC-AGI	84%	#47 of 95	Official leaderboard
HMMT 2026	90.2%	#11 of 19	Lab model card
FrontierMath Tiers 1-3 v2	46.3%	#55 of 86	Official leaderboard
FrontierMath Tier 4 v2	17.1%	#50 of 68	Official leaderboard
CritPt	8.30%	#10 of 13	Lab model card
ProofBench	5.00%	#52 of 59	Official leaderboard
ProofBench v1.1	6.00%	#49 of 52	Official leaderboard

Coding

Benchmark	Score	Rank	Source
SWE-bench Verified	80.2%	#19 of 155	Lab model card
SciCode	48.7%	#12 of 49	Lab model card
SWE-Bench Pro	55.9%	#47 of 81	Lab model card
WebDev Arena	1,407	#65 of 105	arena.ai, via Epoch AI
Code Arena	1,408	#82 of 128	Official leaderboard
Code Migration	13.7%	#59 of 78	Official leaderboard
Vibe Code Bench	19.1%	#92 of 120	Official leaderboard
Image-to-WebDev Arena	1,410	#50 of 55	Official leaderboard
FrontierCode 2.0	3.96%	#30 of 32	Official leaderboard
Code Migration v1.1	10.4%	#29 of 30	Official leaderboard
IOI v2 (Vals)	9.33%	#44 of 45	Official leaderboard
ReactBench	7.69%	#28 of 28	Official leaderboard

Agents and tools

Benchmark	Score	Rank	Source
MCP Atlas	79.2%	#15 of 37	Official leaderboard
BrowseComp	77.4%	#35 of 81	Lab model card
Toolathlon-Verified	54.4%	#18 of 25	Official leaderboard
Terminal-Bench 2.1	64.7%	#41 of 56	Lab model card
τ³-bench Banking	15.5%	#18 of 22	Lab model card
Terminal-Bench 4.0	0.91%	#38 of 40	Official leaderboard
Mystery Game Puzzles	6.00%	#80 of 84	Epoch AI run
Terminal-Bench 4.1	0.91%	#41 of 43	Official leaderboard
Agent Arena	-18.06%	#55 of 56	Official leaderboard

Professional work

Benchmark	Score	Rank	Source
CorpFin v2	69.6%	#6 of 119	Official leaderboard
TaxEval v2	75.5%	#18 of 129	Official leaderboard
MedScribe	84.1%	#37 of 97	Official leaderboard
LegalBench	83%	#62 of 138	Official leaderboard
MortgageTax	62.3%	#52 of 87	Official leaderboard
Harvey LAB (Legal Agent Benchmark)	1.67%	#53 of 80	Official leaderboard
Public Benefits Bench	59.4%	#34 of 50	Official leaderboard
Legal Research Bench	25.5%	#66 of 93	Official leaderboard
MedCode	37.9%	#69 of 96	Official leaderboard
Tax Agent Bench	15.4%	#62 of 84	Official leaderboard
GDPval-AA v2	1,269	#20 of 27	Lab model card
Vals Finance Agent v2	41.3%	#62 of 80	Official leaderboard
Vals Index v2	31.9%	#53 of 65	Official leaderboard
CorpBench Work	66.2	#14 of 17	Official leaderboard
Excel Modeling Benchmark	31.8%	#64 of 76	Official leaderboard
CAD Arena	0.14	#11 of 12	Official leaderboard
Vals Legal Index	57.8%	#36 of 38	Official leaderboard
Harvey LAB v1.1 (HLAB-Vals-1.1)	2.92%	#37 of 38	Official leaderboard
Vals Index v2.1	25.5%	#48 of 49	Official leaderboard
Excel Modeling Benchmark v1.1	40.8%	#42 of 42	Official leaderboard
Vals Finance Agent v2.1	44.8%	#40 of 40	Official leaderboard
Vals Finance Index	36.3%	#40 of 40	Official leaderboard
Vals Index v2.2	27.6%	#37 of 37	Official leaderboard
DeepFinance	10.9%	#23 of 23	Official leaderboard
AA-Briefcase	917	#7 of 7	Lab model card

Vision, audio and robotics

Benchmark	Score	Rank	Source
Audio MC	54.9%	#4 of 18	Official leaderboard
VoiceBench	90.1%	#4 of 12	Lab model card
CharXiv	81.3%	#28 of 75	Lab model card
MMAU	77%	#7 of 15	Lab model card
MMMU-Pro	74%	#43 of 90	Lab model card
ZeroBench	11%	#38 of 76	Official leaderboard
Vision Arena	1,208	#75 of 136	Official leaderboard
Chartography	9.10%	#42 of 43	Official leaderboard

Safety and security

Benchmark	Score	Rank	Source
StrongREJECT	98.4%	#2 of 19	Lab model card
BioTIER Permit	99.8%	#15 of 69	Official leaderboard
Human Pathogen Capabilities Test	54.9%	#36 of 86	Official leaderboard
ABC Bench Screening Evasion	60.2%	#8 of 17	Official leaderboard
Agentic BAIM-LLM Evaluation (ABLE)	57.5%	#9 of 18	Official leaderboard
Molecular Biology Capabilities Test	48.2%	#48 of 90	Official leaderboard
CyberBench (Vals)	68.8%	#14 of 26	Official leaderboard
nocot-bench (NCRI)	70.2	#100 of 185	Official leaderboard
World-Class Bio	44.2%	#41 of 74	Official leaderboard
ABC Bench Fragment Design	88.4%	#15 of 27	Official leaderboard
ABC Bench Liquid-Handling Robot	87.3%	#20 of 34	Official leaderboard
CyberBench v1.1 (Vals)	64.1%	#31 of 44	Official leaderboard
BioTIER Refuse	24.9%	#51 of 69	Official leaderboard
VCT-v2	34.5%	#21 of 25	Official leaderboard

Other

Benchmark	Score	Rank	Source
Vals Multimodal Index	50%	#26 of 33	Official leaderboard
SAGE (Vals)	32.7%	#70 of 81	Official leaderboard
Artificial Analysis Intelligence Index v4.1	40.0	#8 of 8	Lab model card

Its lab's numbers put it at 149.8, in line with its ECI

The Self-Reported Index fits one score to the 21 capability results Thinking Machines reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 147.5 to 152.1. How we compute it.

Its lab's numbers next to 7 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark	Lab says	Board says	Gap
SimpleQA Verified	20.6% SimpleQA Verified	19.1% Epoch AI run, xhigh	+1.5 pts
GPQA Diamond	89.5% GPQA Diamond	88.5% Epoch AI run, xhigh	+1.0 pts
MCP Atlas	79.6% MCP Atlas public 79.6 / all 79.2	79.2% Leaderboard	+0.4 pts
ARC-AGI-2	40.1% ARC-AGI-2	40.1% Leaderboard, xhigh	Same
Audio MC	54.9% Audio MC	54.9% Leaderboard, text output	Same
ARC-AGI	84% ARC-AGI-1	84% Leaderboard, xhigh	Same
Toolathlon-Verified	54.4% Toolathlon Verified	54.4% Leaderboard, xhigh	Same

Thinking Machines's models over time

Thinking Machines's models in color, against every other model on the Epoch Capabilities Index, by release date.

Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

Inkling-Small (Thinking Machines news) Jul 30, 2026

AA Index v4.1 40
SWEBench Verified (bash-only harness) 80.2
SWEBench Pro public 55.9
Terminal Bench 2.1 (best harness) 64.7
SciCode 48.7
GDPval-AA v2 1269
MCP Atlas public 79.6 / all 79.2 79.6
Tau 3 Banking 15.5
BrowseComp with context management 77.4
Toolathlon Verified 54.4
AA-Briefcase 917
GPQA Diamond 89.5
Best of HLE text only 31.6 / HLE with tools 47.8 47.8
AIME 2026 95.5
HMMT Feb 2026 90.2
CritPt 8.3
ARC-AGI-1 84
ARC-AGI-2 40.1
SimpleQA Verified 20.6
AA Omniscience index -9
IFBench 82.2
Global-MMLU-Lite 86.7
StrongREJECT 98.4
FORTRESS adversarial 71.6 (benign 96.9)
MMMU Pro Standard 10 74
Best of Charxiv RQ original 77.4 / with python 81.3 81.3
Audio MC 54.9
MMAU 77
VoiceBench 90.1
Best of ForecastBench Brier Index no search 61.3 / with search 61.5 61.5
Prophet Arena Brier Score (lower is better)

Its system card

Every result in the longer documents the lab published with it.

Inkling-Small Model Card Jul 30, 2026