Thinking MachinesUnited States· released Jul 2026

Inkling

Name: Inkling
Author: Thinking Machines

Ranked #35, 57.2% behind the frontier on average.

Thinking Machines, United States. One of 2 models from this lab on the Index.

Third-party publishedMedium confidence (7+ comparable tests, 3+ Index categories)

Snapshot September 8, 2026

OpenCharts Index

42.8

Ranked #35 (35 to 36) of 36 · 57.2% behind the frontier on average · 15 comparable tests · 3 counting categories

Drop any single test and the Index lands between 36.5 and 46.4, anywhere from #35 to #36. Neighbours inside that range are ties.

Needs more results (3 of 8 so far)

Averaging 26.4 on the 3 it has; the score opens at 4 of 8.

Distance to today's frontier on the hardest tests. Not a measure of intelligence.

What it's known for

Inkling, in 15 results.

Strongest in reasoning (57.3) and coding (38.7). It trails the frontier most in math, 67.6% behind on average. No comparable results yet in agentic, multimodal and long context. Between 50% and 60% behind the best published results, on average. Drop any single test and it would sit anywhere from #35 to #36.

Strongest category

Reasoning

Hard, novel problems: graduate science, abstract puzzles, expert exams.

57.3

GPQA Diamond88.3% · 89% of best
ARC-AGI-179.5% · 81% of best
SimpleBench50.0% · 61% of best
ARC-AGI-236.5% · 39% of best
CritPt5.4% · 17% of best

Mean of 5 of the category's 6 comparable tests. Counts toward the Index.

Best single result

GPQA Diamond

Answer a PhD-level multiple-choice question in physics, chemistry or biology that a web search will not settle.

88.3%

89.4% of the best (95.8%), above chance · #38 of 78.

Expert-level science questions are the closest thing to asking a specialist colleague. A model that gets them right can check a technical claim instead of echoing it.

Where it trails

ProofBench

Write a complete, rigorous proof, graded for rigor rather than for a final answer.

0.0%

100.0% behind the best published result (100.0%) · #61 of 61.

A proof shows the reasoning itself. A right answer with a wrong argument scores nothing.

Not measured yet

No comparable result yet in these 3 categories. A blank is a blank, never a zero, and it does not lower the Index.

Agentic
Multimodal
Long context

Covered, but not averaged into the Index:

Knowledgesingle-test category, shown beside the Index · 53.3
Human preferencesingle-test category, shown beside the Index · 80.7

Every score, with its source

See it at work

What Inkling was asked to do, and how it did.

Every test it has taken, by category, with the kind of task it faced, who produced the number and where the result landed against the best published one. The examples are original and illustrative, never items from the datasets themselves.

Human preference · mean 80.7 across 1 comparable test · a single-test category, shown beside the Index and never averaged into it

Text ArenaLeaderboard

Answer the same prompt as a rival model; real users vote blind on which answer they prefer.

1,439

40% expected win rate against the board leader (1,507)
#58 of 92

board 2026-09-02

Source: Arena

Every result is read against the best published one on its test: percent scores as a share of the best above chance, Arena ratings as the expected win rate against the board leader, open-ended values on a log scale. Sitting a harder exam never lowers a model. Every number links to the publisher that produced it.

Head to head

Inkling against whoever you pick.

Choose a rival. Every counted test both have taken appears side by side as a share of the best published result on that test, with the category means above. A win is a higher share; a gap under 2 points is a tie, because one item on a 45-problem exam is 2.2 points and publishers' own standard errors are of that order.

Inkling

#35·Index 42.8D

0 : 13

wins · 0 ties · 13 shared

GPT-6 Astra

#1·Index 99.0A+

GPT-6 Astra comes out ahead on 13 of the 13 tests both have taken.

Category means

Reasoning
57.3
99.6
Knowledge
53.3
100.0
Coding
38.7
97.7
Math
32.4
99.8
Agentic
—
98.5
Human preference
80.7
—

Shared tests · biggest gaps first

ProofBenchMath
0.0%
99.0%
−99.0
FrontierMath Tier 4Math
4.9%
97.6%
−95.0
CritPtReasoning
5.4%
31.7%
−81.4
Code Arena (WebDev)Coding
1,409
1,797
−80.7
FrontierCodeCoding
14.0%
53.3%
−73.4
WeirdMLCoding
32.3%
92.9%
−65.2
FrontierMath (Tiers 1–3)Math
33.3%
93.7%
−64.4
ARC-AGI-2Reasoning
36.5%
95.0%
−61.5
SimpleQA VerifiedKnowledge
40.3%
75.6%
−46.7
ARC-AGI-1Reasoning
79.5%
98.5%
−19.3
SciCodeCoding
46.1%
56.5%
−16.7
Mock AIME 2024–2025Math
88.9%
100.0%
−11.1
GPQA DiamondReasoning
88.3%
95.8%
−10.6

Bars are distances to the best published result on each test (percent scores above chance, Arena ratings as win rate against the leader, open-ended values on a log scale). Raw values as the publishers report them.

Open GPT-6 Astra

Scores

Every test, every source

The best published run per benchmark, the raw value as the publisher reports it, who produced the number, how close it comes to the best published result on that test, and where the model sits among every model scored on it. Percent scores are read above chance, Arena ratings as a win rate against the board leader, open-ended values on a log scale. A test with too few models to compare against is shown but not counted.

Category means

Reasoning57.3 · 5/6
Knowledgebeside53.3 · 1/1
Coding38.7 · 4/7
Math32.4 · 4/5
Agentic—
Multimodal—
Human preferencebeside80.7 · 1/1
Long context—

Mean · tests taken of the category's comparable tests. A faded category either rests on fewer than half of them or is a single-test category (marked “beside”), which is shown next to the Index and never averaged into it.

GPQA DiamondEpoch-runReasoning

88.3%

89.4% of the best (95.8%), above chance
52th percentile of 78 · n = 198

xhigh

Source: Epoch AI — AI Benchmarking Hub

ARC-AGI-2LeaderboardReasoning

36.5%

38.5% of the best (95.0%)
40th percentile of 51 · n = 120

Source: Epoch AI — AI Benchmarking Hub

ARC-AGI-1LeaderboardReasoning

79.5%

80.7% of the best (98.5%)
38th percentile of 51 · n = 100