Multimodal
Like-for-like- Gemma 4 31B
- 76.9
- Kimi K2.5
- 78.5
- Weighted basis
- 1 vs 1 rows
- Reading
- Kimi K2.5 leads
Gemma 4 31B vs Kimi K2.5
Updated August 18, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.
Decision reading
Gemma 4 31B has the higher public score estimate, 60.15 versus 58.89, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.
7 results are shared. Category rows based on different benchmark sets are marked directional and do not name a winner.
Which one for your work
Recommendations appear only when a shared evidence basis or an explicit operating constraint supports the call. Secondary and unsupported use cases stay disclosed below the initial list.
No workload recommendation clears the current evidence threshold.
Use the matched evidence, workload costs, and sourced specifications below instead of treating a point score as a universal answer.
Show secondary and unsupported calls
Coding work
Code generation, repair, and software-engineering tasks
Not enough matched evidence
The category averages use different weighted benchmark sets, so they are directional rather than like-for-like.
Confidence: limited
Agentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
No shared weighted benchmark basis supports a winner.
Confidence: limited
Long documents
Prompts that approach the documented context limit
No clear pick
The documented context windows are equal.
Confidence: documented
Chat turn cost
1K fresh input + 500 output tokens
Not enough matched evidence
A complete comparable API-rate estimate is not available for both models.
Confidence: listed-rates
Cache-heavy agent loop cost
200K cached + 20K fresh input + 10K output tokens
Not enough matched evidence
A complete comparable API-rate estimate is not available for both models.
Confidence: rate-fallback
Repository review cost
50K fresh input + 3K output tokens
Not enough matched evidence
A complete comparable API-rate estimate is not available for both models.
Confidence: listed-rates
What is actually comparable
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
1 Gemma 4 31B only
7 Shared
38 Kimi K2.5 only
- Shared results
- 7
- Gemma 4 31B only
- 1
- Kimi K2.5 only
- 38
- Like-for-like categories
- 1 / 8
2 categories use different evidence sets. Those rows remain visible for coverage context but do not name a winner.
Category results, on a stated basis
Each row states whether both averages use the same weighted benchmark set. Directional and not-comparable rows remain visible, but they never receive a winner in this template.
Coding
Directional only- Gemma 4 31B
- 41.6
- Kimi K2.5
- 59.4
- Weighted basis
- 1 vs 4 rows
- Reading
- Directional only
Knowledge
Directional only- Gemma 4 31B
- 52.9
- Kimi K2.5
- 56.9
- Weighted basis
- 3 vs 4 rows
- Reading
- Directional only
Agentic
Not comparable- Gemma 4 31B
- Not measured
- Kimi K2.5
- 55.0
- Weighted basis
- 0 vs 2 rows
- Reading
- Not comparable
Reasoning
Not comparable- Gemma 4 31B
- Not measured
- Kimi K2.5
- 61.0
- Weighted basis
- 0 vs 1 rows
- Reading
- Not comparable
Math
Not comparable- Gemma 4 31B
- Not measured
- Kimi K2.5
- 60.6
- Weighted basis
- 0 vs 4 rows
- Reading
- Not comparable
Multilingual
Not comparable- Gemma 4 31B
- Not measured
- Kimi K2.5
- 82.3
- Weighted basis
- 0 vs 1 rows
- Reading
- Not comparable
Instruction following
Not comparable- Gemma 4 31B
- Not measured
- Kimi K2.5
- 93.9
- Weighted basis
- 0 vs 1 rows
- Reading
- Not comparable
| Category | Gemma 4 31B | Kimi K2.5 | Weighted basis | Reading |
|---|---|---|---|---|
| Multimodal | 76.9 | 78.5 | Like-for-like1 vs 1 rows | Kimi K2.5 leads |
| Coding | 41.6 | 59.4 | Directional only1 vs 4 rows | Directional only |
| Knowledge | 52.9 | 56.9 | Directional only3 vs 4 rows | Directional only |
| Agentic | Not measured | 55.0 | Not comparable0 vs 2 rows | Not comparable |
| Reasoning | Not measured | 61.0 | Not comparable0 vs 1 rows | Not comparable |
| Math | Not measured | 60.6 | Not comparable0 vs 4 rows | Not comparable |
| Multilingual | Not measured | 82.3 | Not comparable0 vs 1 rows | Not comparable |
| Instruction following | Not measured | 93.9 | Not comparable0 vs 1 rows | Not comparable |
Shape of the matched evidence
Only shared public evidence is shown. Sparse evidence stays a ruled list rather than being closed into a radar shape.
Too few matched category axes support a radar. The ruled list below shows only shared benchmark results; positions use each benchmark’s normalized display scale when available.
SWE-Rebench
Coding
HLE
Knowledge
GPQA
Knowledge
MMLU-Pro
Knowledge
MMMU-Pro
Multimodal
What each workload costs
Three fixed token mixes turn per-token rates into comparable decisions. Each scenario states context fit and whether cached input had to fall back to the published list-input rate.
Chat turn
1K fresh input + 500 output tokens
- Gemma 4 31B
- Self-hosted; infrastructure cost varies
- Fits in one request
- Kimi K2.5
- $0.0021
- Fits in one request
Gemma 4 31B has no comparable published API token rate.
Repository review
50K fresh input + 3K output tokens
- Gemma 4 31B
- Self-hosted; infrastructure cost varies
- Fits in one request
- Kimi K2.5
- $0.039
- Fits in one request
Gemma 4 31B has no comparable published API token rate.
Cache-heavy agent loop
200K cached + 20K fresh input + 10K output tokens
- Gemma 4 31B
- Self-hosted; infrastructure cost varies
- Fits in one request
- Cached-input rate unavailable
- Kimi K2.5
- $0.162
- Fits in one request
- Cached input priced at the published list-input rate
Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate. Gemma 4 31B has no comparable published API token rate.
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationGemma 4 31BKimi K2.5
Context window
Maximum documented context; output-token limits may be lower.
Gemma 4 31B
Kimi K2.5
256K
API model ID
Gemma 4 31B
gemma-4-31b-it
Google Gemma Gemini API guideKimi K2.5
Not sourced
Cached-input rate
A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
Gemma 4 31B
No comparable hosted API rate
Kimi K2.5
Not published
Documented inputs
Gemma 4 31B
text, image
Google Gemma 4 model documentationKimi K2.5
Not sourced
Documented outputs
Gemma 4 31B
Kimi K2.5
Not sourced
Provider availability
Gemma 4 31B
Generally Available · Gemini API, Google AI Studio, open weights
Google Gemma Gemini API guideKimi K2.5
Not sourced
Reasoning profile
Gemma 4 31B
Reasoning
Kimi K2.5
Non-Reasoning
Weight access
Gemma 4 31B
Open Weight
Kimi K2.5
Open Weight
License
Gemma 4 31B
Open Weight
Kimi K2.5
Open Weight
Release date
Gemma 4 31B
2026-04-02
Kimi K2.5
2026-02-01
If you already use one of these models
- Deployment change
- The models list different providers, so authentication, endpoint behavior, limits, and feature support may change.
- Quality signal
- Gemma 4 31B has the higher public score estimate, 60.15 versus 58.89, but the 90% score intervals overlap.
- Workload cost
- A complete comparable API-rate estimate is not available for both models.
- Context tradeoff
- Both models list 256K.
Run the same representative tasks against both endpoints before changing production traffic.
Self-host vs API cost
Estimates at 50,000 req/day · 1000 tokens/req average.
Gemma 4 31B
API / mo$0
Self-host / mo$429
Break-even—
Kimi K2.5
API / mo$2,700
Self-host / mo$5,221
Break-even132M/day
Benchmark evidence
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Browse raw public benchmark evidence46 rows
Agentic
Gert Labs
Shared sourceGemma 4 31B35.26%Kimi K2.545.88%Kimi K2.5 leads this result
Terminal-Bench 2.0
Gemma 4 31B—Kimi K2.550.8%SourceNot directly comparable
BrowseComp
Gemma 4 31B—Kimi K2.560.6%SourceNot directly comparable
Claw-Eval
Gemma 4 31B—Kimi K2.552.3%SourceNot directly comparable
QwenClawBench
Gemma 4 31B—Kimi K2.554.3%SourceNot directly comparable
τ³-bench results
Gemma 4 31B—Kimi K2.565.7%SourceNot directly comparable
DeepSearchQA
Gemma 4 31B—Kimi K2.577.1%SourceNot directly comparable
DeepPlanning
Gemma 4 31B—Kimi K2.514.4%SourceNot directly comparable
Toolathlon
Gemma 4 31B—Kimi K2.527.8%SourceNot directly comparable
MCP Atlas
Gemma 4 31B—Kimi K2.529.5%SourceNot directly comparable
MCP-Tasks
Gemma 4 31B—Kimi K2.559.1%SourceNot directly comparable
WideResearch
Gemma 4 31B—Kimi K2.572.7%SourceNot directly comparable
ResearchClawBench
Gemma 4 31B—Kimi K2.514.0%SourceNot directly comparable
JobBench
Gemma 4 31B—Kimi K2.58.7%SourceNot directly comparable
Coding
SWE-Rebench
Gemma 4 31B41.6%SourceKimi K2.558.5%SourceKimi K2.5 leads this result
React Native Evals
Shared sourceGemma 4 31B75.2%Kimi K2.577.2%Kimi K2.5 leads this result
SWE-bench Verified
Gemma 4 31B—Kimi K2.576.8%SourceNot directly comparable
SWE-bench Verified*
Gemma 4 31B—Kimi K2.570.8%SourceNot directly comparable
LiveCodeBench v6
Gemma 4 31B—Kimi K2.585.0%SourceNot directly comparable
SWE-bench Pro
Gemma 4 31B—Kimi K2.550.7%SourceNot directly comparable
SWE Multilingual
Gemma 4 31B—Kimi K2.573%SourceNot directly comparable
SciCode
Gemma 4 31B—Kimi K2.548.7%SourceNot directly comparable
Reasoning
LongBench v2
Gemma 4 31B—Kimi K2.561%SourceNot directly comparable
Knowledge
GPQA
Gemma 4 31B84.3%SourceKimi K2.587.6%SourceKimi K2.5 leads this result
MMLU-Pro
Gemma 4 31B85.2%SourceKimi K2.587.1%SourceKimi K2.5 leads this result
HLE
Gemma 4 31B26.5%SourceKimi K2.530.1%SourceKimi K2.5 leads this result
HLE w/o tools
Gemma 4 31B19.5%SourceKimi K2.5—Not directly comparable
GPQA-D
Gemma 4 31B—Kimi K2.587.6%SourceNot directly comparable
SuperGPQA
Gemma 4 31B—Kimi K2.569.2%SourceNot directly comparable
MMLU-Pro (Arcee)
Gemma 4 31B—Kimi K2.587.1%SourceNot directly comparable
Math
AIME 2025
Gemma 4 31B—Kimi K2.596.1%SourceNot directly comparable
AIME26
Gemma 4 31B—Kimi K2.595.8%SourceNot directly comparable
AIME25 (Arcee)
Gemma 4 31B—Kimi K2.596.3%SourceNot directly comparable
HMMT Feb 2025
Gemma 4 31B—Kimi K2.595.4%SourceNot directly comparable
HMMT Nov 2025
Gemma 4 31B—Kimi K2.591.1%SourceNot directly comparable
HMMT Feb 2026
Gemma 4 31B—Kimi K2.587.1%SourceNot directly comparable
MMAnswerBench
Gemma 4 31B—Kimi K2.581.8%SourceNot directly comparable
FrontierMath v2 (Tiers 1-3)
Gemma 4 31B—Kimi K2.527.900%SourceNot directly comparable
FrontierMath v2 (Tier 4)
Gemma 4 31B—Kimi K2.54.200%SourceNot directly comparable
Multilingual
Multimodal
Instruction following
IFEval
Gemma 4 31B—Kimi K2.593.9%SourceNot directly comparable
Frequently asked questions
Which is better, Gemma 4 31B or Kimi K2.5?
Gemma 4 31B has the higher public score estimate, 60.15 versus 58.89, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.
Which is better for coding, Gemma 4 31B or Kimi K2.5?
The current coding averages use different weighted benchmark sets, so BenchLM does not name a winner from them. Read the shared benchmark rows directly and test the models on the same task set.
Which is better for agentic tasks, Gemma 4 31B or Kimi K2.5?
The published evidence does not provide a shared weighted agentic tasks basis for both models, so BenchLM does not name a agentic tasks winner.
Which costs less, Gemma 4 31B or Kimi K2.5?
Both models do not have comparable published API token rates, so this page does not name a universal price winner.
Which has the larger context window, Gemma 4 31B or Kimi K2.5?
Both models list the same context window, 256K.
Related comparisons
Last updated August 18, 2026
Watch Gemma 4 31B vs Kimi K2.5
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.