Reflection AIUnited States· released Oct 2026· open weights

Beam 1.0

Name: Beam 1.0
Author: Reflection AI

Provisional: 8 comparable tests so far. Ranked once it reaches 6 across 2 counting categories.

Reflection AI, United States. One of 1 model from this lab on the Index.

Third-party publishedLow confidence (thin coverage)

Snapshot November 10, 2026

OpenCharts Index

48.4

Provisional: not enough comparable tests or counting categories to rank yet · 8 comparable tests · 0 counting categories

Drop any single test and the Index lands between 42.0 and 56.0.

Needs more results (1 of 5 so far)

Averaging 24.7 on the 1 it has; the score opens at 4 of 5.

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

What it's known for

Beam 1.0, in 8 results.

Its widest gap is on ProofBench, 82.0% behind the best published result. No comparable results yet in knowledge and multimodal. Between 50% and 60% 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

ARC-AGI-1

Solve the original Abstraction and Reasoning Corpus: small grids, a handful of examples, one hidden rule.

74.0%

75.1% of the best (98.5%) · #50 of 67.

The first ARC set is close to saturated at the frontier, so it now shows whether smaller models can generalize at all.

Where it trails

ProofBench

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

18.0%

82.0% behind the best published result (100.0%) · #55 of 81.

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

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
Multimodal

Covered, but not averaged into the Index:

Reasoning3 of 8 tests · needs 4 · mean 45.4
Coding2 of 6 tests · needs 3 · mean 46.4
Math1 of 4 tests · needs 2 · mean 18.0
Agentic1 of 6 tests · needs 3 · mean 58.0
Human preferencesingle-test category, shown beside the Index · 74.3

Every score, with its source

See it at work

What Beam 1.0 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 74.3 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.

1,435

37% expected win rate against the board leader (1,526)
#80 of 113

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

Beam 1.0 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.

Beam 1.0

provisional·Index 48.4D

0 : 8

wins · 0 ties · 8 shared

GPT-6 Astra

#1·Index 93.7A

GPT-6 Astra comes out ahead on 8 of the 8 tests both have taken.

Category means

Reasoning
45.4
96.1
Knowledge
—
80.4
Coding
46.4
93.4
Math
18.0
98.8
Agentic
58.0
90.6
Multimodal
—
89.6
Human preference
74.3
85.2

Shared tests · biggest gaps first

ProofBenchMath
18.0%
99.0%
−81.0
ARC-AGI-2Reasoning
23.5%
95.0%
−75.3
Code Arena (WebDev)Coding
1,438
1,767
−65.3
CritPtReasoning
13.1%
31.7%
−51.6
ARC-AGI-1Reasoning
74.0%
98.5%
−24.9
APEX-AgentsAgentic
48.8%
64.7%
−18.8
SciCodeCoding
48.5%
56.5%
−11.9
Text ArenaHuman preference
1,435
1,474
−10.9

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

Reasoning45.4 · 3/8
Knowledge—
Coding46.4 · 2/6
Math18.0 · 1/4
Agentic58.0 · 1/6
Multimodal—
Human preferencebeside74.3 · 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.

ARC-AGI-2LeaderboardReasoning

23.5%

24.7% of the best (95.0%)
27th percentile of 67 · n = 120

xhigh

Source: Epoch AI — AI Benchmarking Hub

ARC-AGI-1LeaderboardReasoning

74.0%

75.1% of the best (98.5%)
26th percentile of 67 · n = 100

xhigh

Source: Epoch AI — AI Benchmarking Hub

CritPtLeaderboardReasoning

13.1%