StepFunChina· released Sep 2026

Step 5 Preview

Name: Step 5 Preview
Author: StepFun

Provisional: 6 comparable tests so far. Ranked once it reaches 6 across 2 counting categories.

StepFun, China. One of 1 model from this lab on the Index.

Third-party publishedLow confidence (thin coverage)

Snapshot November 10, 2026

OpenCharts Index

64.7

Provisional: not enough comparable tests or counting categories to rank yet · 6 comparable tests · 0 counting categories

Drop any single test and the Index lands between 59.8 and 70.4.

Needs more results (0 of 5 so far)

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

What it's known for

Step 5 Preview, in 6 results.

Its widest gap is on Code Arena (WebDev), 60.8% behind the best published result. No comparable results yet in knowledge and agentic. Between 35% and 40% behind the best published results, on average.

Strongest category

No category counts yet: a category counts once it has more than one comparable test and the model has taken at least half of them. The scores it has are below.

Best single result

SciCode

Write research-grade scientific code, one sub-problem at a time, that passes unit tests.

58.9%

88.1% of the best (66.9%) · #14 of 96.

Scientific programming needs the math and the code to be right at the same time.

Where it trails

Code Arena (WebDev)

Build a web app from the same prompt as a rival model; real users vote blind on the result.

1,572

60.8 points short of parity with the board leader (1,818) · #37 of 110.

People judging finished apps side by side is the most honest measure of front-end quality.

Not measured yet

No comparable result yet in these 2 categories. A blank is a blank, never a zero, and it does not lower the Index.

Knowledge
Agentic

Covered, but not averaged into the Index:

Reasoning1 of 8 tests · needs 4 · mean 57.9
Coding2 of 6 tests · needs 3 · mean 63.7
Math1 of 4 tests · needs 2 · mean 42.0
Multimodal1 of 4 tests · needs 2 · mean 81.8
Human preferencesingle-test category, shown beside the Index · 78.0

Every score, with its source

See it at work

What Step 5 Preview 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.

Multimodal · mean 81.8 across 1 comparable test · not counted toward the Index yet (fewer than half the category's tests)

Vision ArenaLeaderboard

Answer prompts that include an image; real users vote blind between two models' answers.

1,245

41% expected win rate against the board leader (1,309)
#50 of 71

preview · board 2026-11-09

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

Step 5 Preview 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.

Step 5 Preview

provisional·Index 64.7C

1 : 5

wins · 0 ties · 6 shared

GPT-6 Astra

#1·Index 93.7A

GPT-6 Astra comes out ahead on 5 of the 6 tests both have taken.

Category means

Reasoning
57.9
96.1
Knowledge
—
80.4
Coding
63.7
93.4
Math
42.0
98.8
Agentic
—
90.6
Multimodal
81.8
89.6
Human preference
78.0
85.2

Shared tests · biggest gaps first

ProofBenchMath
42.0%
99.0%
−57.0
Code Arena (WebDev)Coding
1,572
1,767
−46.3
CritPtReasoning
20.9%
31.7%
−30.2
Vision ArenaMultimodal
1,245
1,283
−10.7
Text ArenaHuman preference
1,448
1,474
−7.2
SciCodeCoding
58.9%
56.5%
+3.7

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, and so is an archived test: one whose publisher no longer covers new models.

Category means

Reasoning57.9 · 1/8
Knowledge—
Coding63.7 · 2/6
Math42.0 · 1/4
Agentic—
Multimodal81.8 · 1/4
Human preferencebeside78.0 · 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.

CritPtLeaderboardReasoning

20.9%

57.9% of the best (36.0%)
73th percentile of 102

preview

Source: Epoch AI — AI Benchmarking Hub

SciCodeLeaderboardCoding

58.9%

88.1% of the best (66.9%)
86th percentile of 96 · n = 338

preview

Source: Epoch AI — AI Benchmarking Hub

Code Arena (WebDev)LeaderboardCoding

1,572