BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases8

Gemini 4 Carbon vs Kimi K2.5

Full profiles: Gemini 4 Carbon (Google) and Kimi K2.5 (Moonshot AI): every benchmark, price and rank.

Updated · Data from 9 sources

As of 29 Oct 2026, Gemini 4 Carbon is ahead of Kimi K2.5 on quality (65.1 vs 58.8 on the BenchLeader Index), but Kimi K2.5 is 6.7× cheaper at $1.20 vs $8.00 per 1M tokens blended. Gemini 4 Carbon leads in every category we measure both in.

Key facts (as of 29 Oct 2026)

  • Gemini 4 Carbon scores 65.1 on the BenchLeader Index and Kimi K2.5 scores 58.8 (BenchLeader, 29 Oct 2026).
  • Gemini 4 Carbon costs $8.00 per 1M tokens blended (list price, 29 Oct 2026).
  • Kimi K2.5 costs $1.20 per 1M tokens blended (list price, 29 Oct 2026).
  • Of the 6 benchmarks both models have, Gemini 4 Carbon leads 6 (BenchLeader, 29 Oct 2026).
Gemini 4 Carbon benchmarks high Google Kimi K2.5 benchmarks thinking Moonshot AIOpen in the comparison tool

Which to pick for…

The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.

Use casePickWhyConfidence
CodingGemini 4 Carbon72 vs 56 category score; Vibe Code Bench v1.1 89.1% vs 17.5%High
ReasoningGemini 4 Carbon79 vs 67 category score; Humanity's Last Exam (AA) 60.4% vs 30.7%High
KnowledgeGemini 4 Carbon76 vs 60 category scoreHigh
Long documentsEither65 vs 64 category score; AA-LCR v1.1 79.3% vs 78.0%Either
Chat quality (human preference)–Only Kimi K2.5 has an Arena rating (63 category score)Not comparable
Lowest costKimi K2.5$1.20 vs $8.00 per 1M blended; agentic coding session $0.048 vs $0.320–
Fastest first answerKimi K2.535.4 s (Kimi K2.5 (thinking)) vs 103.1 s (Gemini 4 Carbon (high)) to the first answer token–
High volume on a budgetNeither flagshipA smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs–

Key differences

  • Gemini 4 Carbon (high) leads on quality: 65.1 vs 58.8.
  • Gemini 4 Carbon (high) is stronger in coding, knowledge, long context, reasoning.
  • Kimi K2.5 (thinking) is stronger in agents & tools, human preference, instruction following, maths, multimodal.
  • Kimi K2.5 (thinking) is 6.7× cheaper ($1.20 vs $8.00 per 1M blended).

At a glance

SpecGemini 4 CarbonKimi K2.5
DeveloperGoogleMoonshot AI
Released28 Oct 202627 Jan 2026
Context window (tokens)1,000,000262,144
Compared atGemini 4 Carbon (high)Kimi K2.5 (thinking)
Open weightsNoYes
Price in / out / cached input per 1M$4.00 / $20.00 / –$0.600 / $3.00 / $0.100

Category by category

Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.

CategoryGemini 4 Carbon (high)Kimi K2.5 (thinking)Leader
Reasoning78.967.3Gemini 4 Carbon
Coding71.755.7Gemini 4 Carbon
Agents & tools–45.1not comparable
Maths–56.2not comparable
Knowledge75.860.0Gemini 4 Carbon
Instruction following–69.0not comparable
Human preference–63.4not comparable
Multimodal–59.3not comparable
Long context64.563.8level

Where sources disagree

Across the 6 benchmarks both have, Gemini 4 Carbon leads 6, Kimi K2.5 leads 0. Every overall index that rates both puts Gemini 4 Carbon ahead.

Overall indexGemini 4 Carbon (high)Kimi K2.5 (thinking)Leader
BenchLeader Index65.158.8Gemini 4 Carbon

Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).

Price, speed and latency

List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page; suite cost is what Artificial Analysis paid to run its whole Intelligence Index.

MeasureGemini 4 Carbon (high)Kimi K2.5 (thinking)
Blended price per 1M tokens$8.00$1.20
Input / output per 1M$4.00 / $20.00$0.600 / $3.00
Cached input per 1M–$0.100
Price per task: chat reply$0.0076$0.0011
Price per task: summarize a 30-page report$0.060$0.0090
Price per task: code edit$0.054$0.0081
Price per task: agentic coding session$0.320$0.048
Price per task: structured extraction$0.012$0.0018
Cost to run a full benchmark suite$7.76–
Output speed100 tok/s89 tok/s
Time to first answer103.1 s35.4 s
Time to a complete answer108.1 s41.0 s

Each model is compared at its best-ranked reasoning-effort setting. Gemini 4 Carbon (high) scores 65.1 and takes 103 s to its first answer token. Kimi K2.5 (thinking) scores 58.8 and takes 35 s to its first answer token; Kimi K2.5 scores 54.3 in 1 s. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.

Every benchmark

103 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.

MetricPublisherGemini 4 Carbon (high)Kimi K2.5 (thinking)
BenchLeader IndexBenchLeader65.158.8
Coding scoreBenchLeader71.755.7
Knowledge scoreBenchLeader75.860.0
Long context scoreBenchLeader64.563.8
Reasoning scoreBenchLeader78.967.3
Agents & tools scoreBenchLeader–45.1
Human preference scoreBenchLeader–63.4
Instruction following scoreBenchLeader–69.0
Maths scoreBenchLeader–56.2
Multimodal scoreBenchLeader–59.3
Blended price $/MBenchLeader$8.00$1.20
Output speedBenchLeader100 tok/s89 tok/s
Time to first answerBenchLeader103.1 s35.4 s
Context windowBenchLeader1,000,000 tokens262,144 tokens
Reasoning
GPQA DiamondEpoch AI Benchmarking Hub–87.6%†
Humanity's Last ExamScale AI / CAIS–24.4%†
SimpleBenchSimpleBench–46.8%†
Arena Hard PromptsArena (LMArena)–1471
GPQA Diamond (AA)Artificial Analysis–87.9%
Humanity's Last Exam (AA)Artificial Analysis60.4%30.7%
HLE-Diamond (AA)Artificial Analysis69.1%–
GPQA Diamond (Vals)Vals AI–84.1%
Kagi LLM BenchmarkKagi LLM Benchmark–78.5%
ARC-AGI-1ARC Prize–65.3%†
ARC-AGI-2ARC Prize–11.8%†
CritPtArtificial Analysis36.0%3.1%
CritPt v2 hard subset (AA)Artificial Analysis63.0%–
CL-benchEpoch AI Benchmarking Hub–19.3%†
CL-bench LifeEpoch AI Benchmarking Hub–13.2%†
Coding
SWE-bench Verified (Epoch)Epoch AI Benchmarking Hub–73.8%†
SciCodeSciCode64.2%49.0%†
WeirdMLWeirdML–45.6%†
Arena CodingArena (LMArena)–1502
Arena WebDevArena (LMArena)–1437
SciCode (AA)Artificial Analysis64.2%–
LiveCodeBenchVals AI–83.9%
SWE-bench (Vals)Vals AI–70.0%
SWE-bench Verified (bash only)SWE-bench–70.8%†
SWE-bench Verified (any scaffold)SWE-bench–70.8%†
Show all 103 benchmarks
MetricPublisherGemini 4 Carbon (high)Kimi K2.5 (thinking)
Coding
Code MigrationVals AI–7.0%
Vibe Code Bench v1.1Vals AI89.1%17.5%
SWE Atlas: Codebase QnAScale AI SEAL–13.1%†
SWE Atlas: RefactoringScale AI SEAL–20.9%†
SWE Atlas: Test WritingScale AI SEAL–25.8%†
ALE-BenchEpoch AI Benchmarking Hub–821.6†
Terminal-Bench 4.0 (AA)Artificial Analysis65.2%–
Terminal-Bench 2.1 (AA)Artificial Analysis–45.7%
FrontierCode 1.1 (AA)Artificial Analysis48.5%–
FrontierCode 2.0 (AA)Artificial Analysis40.6%–
Agents & tools
Terminal-BenchTerminal-Bench–43.2%†
Terminal-Bench HardArtificial Analysis–34.9%
τ²-Bench Telecom (AA)Artificial Analysis–95.9%
Terminal-Bench 2.1 (Vals)Vals AI–42.0%
MCP AtlasScale AI SEAL–64.4%†
τ³-Banking (AA)Artificial Analysis–14.2%
APEX-Agents (AA)Artificial Analysis–11.5%
Finance Agent v2Vals AI–35.8%
Finance Agent v2.1Vals AI67.2%–
Harvey's Legal Agent Benchmark v1.1Vals AI17.5%–
Harvey's Legal Agent Benchmark v1Vals AI–0.0%
Legal Research BenchVals AI55.8%15.9%
Tax Agent BenchVals AI48.9%–
Terminal-Bench 2.0 (Vals)Vals AI–40.5%
Terminal-Bench 4.0 (Vals)Vals AI63.1%–
Vending-Bench 2Andon Labs16484.2†1198.5†
AutomationBenchArtificial Analysis77.8%–
GDP.pdf (AA)Artificial Analysis28.4%–
AA-Briefcase v1.1Artificial Analysis1609–
Terminal-Bench-Science 0.1 (AA)Artificial Analysis64.8%–
Maths
OTIS Mock AIMEEpoch AI Benchmarking Hub–92.2%†
AIME (Vals)Vals AI–95.6%
AIME 2026MathArena–95.8%
HMMT February 2026MathArena–87.1%
MathArena ApexMathArena–8.8%
Arena MathsArena (LMArena)–1471
Knowledge
SimpleQA VerifiedEpoch AI Benchmarking Hub–34.3%†
AA-Omniscience v1.1Artificial Analysis53.5–
MMLU-ProVals AI–85.9%
CorpFinVals AI–68.3%
TaxEvalVals AI–74.2%
MedQAVals AI–94.4%
PRBench FinanceScale AI SEAL–46.5%†
PRBench LegalScale AI SEAL–43.8%†
MultiNRCScale AI SEAL–35.2%†
CaseLaw v2Vals AI–58.7%
Excel Modeling BenchmarkVals AI–28.5%
Excel Modeling Benchmark v1.1Vals AI76.5%–
MedCodeVals AI–39.3%
MedScribeVals AI–76.4%
AA-Omniscience v1.1: accuracyArtificial Analysis58.3%–
AA-Omniscience v1.1: non-hallucinationArtificial Analysis88.6%–
Instruction following
IFBenchArtificial Analysis–70.2%
MultiChallengeScale AI SEAL–61.4%†
TutorBenchScale AI SEAL–54.6%†
Arena Instruction FollowingArena (LMArena)–1439
Human preference
Arena TextArena (LMArena)–1450
Arena Creative WritingArena (LMArena)–1423
Arena Multi-turnArena (LMArena)–1452
Arena Longer QueriesArena (LMArena)–1458
Multimodal
Arena VisionArena (LMArena)–1269
MMMU-ProArtificial Analysis–75.4%
VISTAScale AI SEAL–41.9%†
MMMU-Pro (Vals)Vals AI–84.3%
MortgageTaxVals AI–66.5%
Blueprint-Bench 2Epoch AI Benchmarking Hub58.6%†–
Long context
Fiction.LiveBench 120kFiction.live–78.1%†
AA-LCR v1.1Artificial Analysis79.3%78.0%
Arena DocumentArena (LMArena)–1436
Composite
Epoch Capabilities IndexEpoch AI Benchmarking Hub–148.1†
AA Intelligence Index v5.0Artificial Analysis59.4–
AA Intelligence Index v4.3.2Artificial Analysis–23.5
AA Progress Index v1Artificial Analysis351.6270.5
Vals IndexVals AI70.4–
Vals Finance IndexVals AI66.3%–
Vals RSI IndexVals AI–6.8%†
Safety & honesty
FORTRESSScale AI SEAL–41.1%†
MASKScale AI SEAL–70.5%†

Data as of 2026-10-29. Best configuration of each model; every score links to its source on the model pages.

What changed

  • Kimi K2.5 (thinking): 1 new result (AA Progress Index v1).
  • Kimi K2.5 (thinking): 1 new result (Harvey's Legal Agent Benchmark v1).
  • Kimi K2.5 (thinking): output price moved from $2.75/M to $3.00/M (9% up).
  • Kimi K2.5 (thinking): BenchLeader Index 59.5 → 59.0.
  • Kimi K2.5 (thinking): 1 score revision.

Rankings: Best LLM overall · Best LLM for coding · Best LLM for agents · Best cheap LLMs

Gemini 4 Carbon vs Kimi K2.5: questions

Is Gemini 4 Carbon better than Kimi K2.5?
Gemini 4 Carbon (high) leads on quality: 65.1 vs 58.8. The BenchLeader Index combines every independent quality benchmark; Gemini 4 Carbon (high) is ahead overall as of 29 Oct 2026, but check the category scores for your use.
Gemini 4 Carbon vs Kimi K2.5: which is cheaper, and will that change?
Kimi K2.5 is cheaper today: $1.20 against $8.00 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
Is Gemini 4 Carbon or Kimi K2.5 better for coding?
Gemini 4 Carbon scores higher in coding (72 vs 56 on the category index, where 50 is average).
Which is faster, Gemini 4 Carbon or Kimi K2.5?
Gemini 4 Carbon streams faster (100 vs 89 output tokens per second, Artificial Analysis 7-day median).
Which is better for long documents?
They are level on long-context benchmarks (65 vs 64). Gemini 4 Carbon accepts more context: 1,000,000 against 262,144 tokens for Kimi K2.5.
More questions (3)
Which has lower latency?
Kimi K2.5 answers first: 35.4 s against 103.1 s to the first answer token, as Artificial Analysis measures it. Reasoning effort drives most of this: a higher setting thinks longer before it answers.
Do independent benchmarks agree on Gemini 4 Carbon vs Kimi K2.5?
Across the 6 benchmarks both have, Gemini 4 Carbon leads 6, Kimi K2.5 leads 0. Every overall index that rates both puts Gemini 4 Carbon ahead.
When was Gemini 4 Carbon released?
Gemini 4 Carbon was released on 28 Oct 2026.