Model vs model

gemma-4-31b vs A.X-K2

A debate-ready pair page: current winner, strongest alternative, decisive benchmarks, and the warning that should travel with the claim.

Use case · Everyday chatbot
Winner · gemma-4-31b
Sources · All public sources

Winner

gemma-4-31b

Google

8benchmarks won

Multilingual

Versus · Everyday chatbot

gemma-4-31b leads this compare set for everyday chatbot.

gemma-4-31b8 of 8 wins

Verified and stable

0 shared benchmarks are still too close to call, so the win stays conditional. This compare uses all public sources, with provider-official evidence labeled separately.

Close calls: 0
Sources: All public

Full compare workspace Disagreement page

100%win share

Challenger

A.X-K2

Unknown

0benchmarks won

Chat / text
Reasoning / math / science

The cases in full

gemma-4-31b case

Multilingual

A.X-K2 case

Chat / text
Reasoning / math / science

What changes the outcome

gemma-4-31b: 101 visible benchmark gaps still leave room for the result to move.
A.X-K2: 124 visible benchmark gaps still leave room for the result to move.

Why this result is surprising

The visible shared evidence is more decisive than usual for this compare set.
Very few shared benchmarks are decisively separating these models.

Why this is not a clean win

0 shared benchmarks are still too close to call, so the win stays conditional. This compare uses all public sources, with provider-official evidence labeled separately.
A.X-K2 remains the strongest alternative once you change use case, mode, or missing-evidence assumptions.

Open full compare workspace Open evidence page Open disagreement page

Decisive benchmarks

8 of 132 benchmarks


Intelligence Index AA · index Text · Chat / text	1458.9% exact aliasverified runtime Row details Raw value 14 Percentile 58.9% Last updated recent Eligibility headline eligible Identity provider alias (0.94) Source label Gemma 4 31B (Non-reasoning) Source row	23n/a exact aliasverified runtimeContext only Row details Raw value 23 Percentile n/a Last updated recent Eligibility benchmark_derived_model Identity provider alias (0.94) Source label A.X-K2 Source row	n/a
GPQA AA · % Text · Reasoning / math / science	76.3%65.9% exact aliasverified runtime Row details Raw value 76.3% Percentile 65.9% Last updated recent Eligibility headline eligible Identity provider alias (0.94) Source label Gemma 4 31B (Non-reasoning) Source row	85.7%n/a exact aliasverified runtimeContext only Row details Raw value 85.7% Percentile n/a Last updated recent Eligibility benchmark_derived_model Identity provider alias (0.94) Source label A.X-K2 Source row	n/a
Humanity's Last Exam AA · % Text · Reasoning / math / science	11.8%61.6% exact aliasverified runtime Row details Raw value 11.8% Percentile 61.6% Last updated recent Eligibility headline eligible Identity provider alias (0.94) Source label Gemma 4 31B (Non-reasoning) Source row	29.6%n/a exact aliasverified runtimeContext only Row details Raw value 29.6% Percentile n/a Last updated recent Eligibility benchmark_derived_model Identity provider alias (0.94) Source label A.X-K2 Source row	n/a
CritPt AA · % Text · Reasoning / math / science	0%50.1% exact aliasverified runtime Row details Raw value 0% Percentile 50.1% Last updated recent Eligibility headline eligible Identity provider alias (0.94) Source label Gemma 4 31B (Non-reasoning) Source row	8.9%n/a exact aliasverified runtimeContext only Row details Raw value 8.9% Percentile n/a Last updated recent Eligibility benchmark_derived_model Identity provider alias (0.94) Source label A.X-K2 Source row	n/a
SciCode AA · % Code · Coding	45.5%45.5% exact aliasverified runtime Row details Raw value 45.5% Percentile 45.5% Last updated recent Eligibility headline eligible Identity provider alias (0.94) Source label Gemma 4 31B (Reasoning) Source row	41%n/a exact aliasverified runtimeContext only Row details Raw value 41% Percentile n/a Last updated recent Eligibility benchmark_derived_model Identity provider alias (0.94) Source label A.X-K2 Source row	n/a
AA-Omniscience accuracy AA · % Text · Chat / text	16.6%35.8% exact aliasverified runtime Row details Raw value 16.6% Percentile 35.8% Last updated recent Eligibility headline eligible Identity provider alias (0.94) Source label Gemma 4 31B (Non-reasoning) Source row	18.6%n/a exact aliasverified runtimeContext only Row details Raw value 18.6% Percentile n/a Last updated recent Eligibility benchmark_derived_model Identity provider alias (0.94) Source label A.X-K2 Source row	n/a
AA-Omniscience non-hallucination AA · % Text · Chat / text	15%41.7% exact aliasverified runtime Row details Raw value 15% Percentile 41.7% Last updated recent Eligibility headline eligible Identity provider alias (0.94) Source label Gemma 4 31B (Reasoning) Source row	67%n/a exact aliasverified runtimeContext only Row details Raw value 67% Percentile n/a Last updated recent Eligibility benchmark_derived_model Identity provider alias (0.94) Source label A.X-K2 Source row	n/a
Long Context Reasoning AA · % Document · Long context	46.7%47.9% exact aliasverified runtime Row details Raw value 46.7% Percentile 47.9% Last updated recent Eligibility headline eligible Identity provider alias (0.94) Source label Gemma 4 31B (Non-reasoning) Source row	70%n/a exact aliasverified runtimeContext only Row details Raw value 70% Percentile n/a Last updated recent Eligibility benchmark_derived_model Identity provider alias (0.94) Source label A.X-K2 Source row	n/a

1-8 of 8

Page 1 of 1Page size

Intelligence Index

AA · index

Text · Chat / text

1458.9%

exact aliasverified runtime

Row details

Raw value: 14
Percentile: 58.9%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.94)
Source label: Gemma 4 31B (Non-reasoning)

Source row

23n/a

exact aliasverified runtimeContext only

Row details

Raw value: 23
Percentile: n/a
Last updated: recent
Eligibility: benchmark_derived_model
Identity: provider alias (0.94)
Source label: A.X-K2

Source row

n/a

GPQA

AA · %

Text · Reasoning / math / science

76.3%65.9%

exact aliasverified runtime

Row details

Raw value: 76.3%
Percentile: 65.9%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.94)
Source label: Gemma 4 31B (Non-reasoning)

Source row

85.7%n/a

exact aliasverified runtimeContext only

Row details

Raw value: 85.7%
Percentile: n/a
Last updated: recent
Eligibility: benchmark_derived_model
Identity: provider alias (0.94)
Source label: A.X-K2

Source row

n/a

Humanity's Last Exam

AA · %

Text · Reasoning / math / science

11.8%61.6%

exact aliasverified runtime

Row details

Raw value: 11.8%
Percentile: 61.6%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.94)
Source label: Gemma 4 31B (Non-reasoning)

Source row

29.6%n/a

exact aliasverified runtimeContext only

Row details

Raw value: 29.6%
Percentile: n/a
Last updated: recent
Eligibility: benchmark_derived_model
Identity: provider alias (0.94)
Source label: A.X-K2

Source row

n/a

CritPt

AA · %

Text · Reasoning / math / science

0%50.1%

exact aliasverified runtime

Row details

Raw value: 0%
Percentile: 50.1%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.94)
Source label: Gemma 4 31B (Non-reasoning)

Source row

8.9%n/a

exact aliasverified runtimeContext only

Row details

Raw value: 8.9%
Percentile: n/a
Last updated: recent
Eligibility: benchmark_derived_model
Identity: provider alias (0.94)
Source label: A.X-K2

Source row

n/a

SciCode

AA · %

Code · Coding

45.5%45.5%

exact aliasverified runtime

Row details

Raw value: 45.5%
Percentile: 45.5%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.94)
Source label: Gemma 4 31B (Reasoning)

Source row

41%n/a

exact aliasverified runtimeContext only

Row details

Raw value: 41%
Percentile: n/a
Last updated: recent
Eligibility: benchmark_derived_model
Identity: provider alias (0.94)
Source label: A.X-K2

Source row

n/a

AA-Omniscience accuracy

AA · %

Text · Chat / text

16.6%35.8%

exact aliasverified runtime

Row details

Raw value: 16.6%
Percentile: 35.8%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.94)
Source label: Gemma 4 31B (Non-reasoning)

Source row

18.6%n/a

exact aliasverified runtimeContext only

Row details

Raw value: 18.6%
Percentile: n/a
Last updated: recent
Eligibility: benchmark_derived_model
Identity: provider alias (0.94)
Source label: A.X-K2

Source row

n/a

AA-Omniscience non-hallucination

AA · %

Text · Chat / text

15%41.7%

exact aliasverified runtime

Row details

Raw value: 15%
Percentile: 41.7%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.94)
Source label: Gemma 4 31B (Reasoning)

Source row

67%n/a

exact aliasverified runtimeContext only

Row details

Raw value: 67%
Percentile: n/a
Last updated: recent
Eligibility: benchmark_derived_model
Identity: provider alias (0.94)
Source label: A.X-K2

Source row

n/a

Long Context Reasoning

AA · %

Document · Long context

46.7%47.9%

exact aliasverified runtime

Row details

Raw value: 46.7%
Percentile: 47.9%
Last updated: recent
Eligibility: headline eligible
Identity: provider alias (0.94)
Source label: Gemma 4 31B (Non-reasoning)

Source row

70%n/a

exact aliasverified runtimeContext only

Row details

Raw value: 70%
Percentile: n/a
Last updated: recent
Eligibility: benchmark_derived_model
Identity: provider alias (0.94)
Source label: A.X-K2

Source row

n/a

gemma-4-31b vs A.X-K2

gemma-4-31b

gemma-4-31b leads this compare set for everyday chatbot.

A.X-K2

gemma-4-31b case

A.X-K2 case

What changes the outcome

Why this result is surprising

Why this is not a clean win

Share the recommendation with the source data attached.

Open or copy the stable surfaces

Use the exact public framing

Pick the voice before you post

Compose a post that keeps the warning attached

Decisive benchmarks

Loading model comparison.

gemma-4-31b vs A.X-K2

gemma-4-31b

gemma-4-31b leads this compare set for everyday chatbot.

A.X-K2

gemma-4-31b case

A.X-K2 case

What changes the outcome

Why this result is surprising

Why this is not a clean win

Share the recommendation with the source data attached.

Open or copy the stable surfaces

Use the exact public framing

Pick the voice before you post

Compose a post that keeps the warning attached

Decisive benchmarks