BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9

GPT-5.6 Terra vs Kimi K3

Full profiles: GPT-5.6 Terra (OpenAI) and Kimi K3 (Moonshot AI): every benchmark, price and rank.

Updated · Data from 11 sources

As of 25 Oct 2026, GPT-5.6 Terra (max) and Kimi K3 (max) are neck and neck on quality (64.4 vs 64.8 on the BenchLeader Index, inside the ±6.0 margin of error), and GPT-5.6 Terra (max) costs 1.3× less, which makes it the better buy for most workloads. Pick GPT-5.6 Terra for maths (69 vs 63); pick Kimi K3 for knowledge (63 vs 57).

GPT-5.6 Terra benchmarks max OpenAI Kimi K3 benchmarks max 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
CodingEither65 vs 65 category score; SciCode 55.0% vs 59.5%Either
Agents & tool useEither59 vs 59 category score; τ³-Banking (AA) 40.2% vs 46.0%Either
MathsGPT-5.6 Terra69 vs 63 category score; FrontierMath Tiers 1–3 86.0% vs 72.2%Medium
ReasoningGPT-5.6 Terra70 vs 66 category score; GPQA Diamond 93.3% vs 93.1%Medium
KnowledgeKimi K363 vs 57 category score; SimpleQA Verified 50.6% vs 43.2%Medium
Long documentsKimi K370 vs 67 category score; AA-LCR 88.7% vs 83.0%Medium
Images and other mediaEither64 vs 64 category score; MMMU-Pro 80.7% vs 80.5%Either
Chat quality (human preference)–Only Kimi K3 has an Arena rating (68 category score)Not comparable
Lowest costGPT-5.6 Terra$4.50 vs $6.00 per 1M blended; agentic coding session $0.168 vs $0.240–
Fastest first answerKimi K352.6 s (Kimi K3 (max)) vs 101.0 s (GPT-5.6 Terra (max)) 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

  • GPT-5.6 Terra (max) and Kimi K3 (max) are neck and neck on quality (64.4 vs 64.8, within the ±6.0 margin).
  • GPT-5.6 Terra (max) is stronger in coding, instruction following, maths, multimodal, reasoning.
  • Kimi K3 (max) is stronger in agents & tools, composite, knowledge, long context, human preference.
  • GPT-5.6 Terra (max) is 1.3× cheaper ($4.50 vs $6.00 per 1M blended).
  • GPT-5.6 Terra (max) streams 3.1× faster (126 vs 41 tokens per second).

At a glance

SpecGPT-5.6 TerraKimi K3
DeveloperOpenAIMoonshot AI
Released9 Jul 202616 Jul 2026
Context window (tokens)1,050,0001,048,576
Compared atGPT-5.6 Terra (max)Kimi K3 (max)
Open weightsNoYes
Price in / out / cached input per 1M$2.00 / $12.00 / $0.200$3.00 / $15.00 / $0.300

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.

CategoryGPT-5.6 Terra (max)Kimi K3 (max)Leader
Reasoning70.166.4GPT-5.6 Terra
Coding65.164.6level
Agents & tools58.959.4level
Maths68.963.4GPT-5.6 Terra
Knowledge57.262.8Kimi K3
Instruction following69.9–not comparable
Human preference–68.1not comparable
Multimodal63.763.6level
Long context66.669.6Kimi K3
Composite76.678.3Kimi K3

Where sources disagree

Across the 47 benchmarks both have, GPT-5.6 Terra leads 26, Kimi K3 leads 21. Every overall index that rates both puts Kimi K3 ahead.

Overall indexGPT-5.6 Terra (max)Kimi K3 (max)Leader
BenchLeader Index64.464.8neck and neck
AA Intelligence Index v4.3.242.143.6Kimi K3
Arena Text1467†1487Kimi K3

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.

MeasureGPT-5.6 Terra (max)Kimi K3 (max)
Blended price per 1M tokens$4.50$6.00
Input / output per 1M$2.00 / $12.00$3.00 / $15.00
Cached input per 1M$0.200$0.300
Price per task: chat reply$0.0044$0.0057
Price per task: summarize a 30-page report$0.031$0.045
Price per task: code edit$0.030$0.041
Price per task: agentic coding session$0.168$0.240
Price per task: structured extraction$0.0064$0.0090
Cost to run a full benchmark suite$1.40$2.00
Output speed126 tok/s41 tok/s
Time to first answer101.0 s52.6 s
Time to a complete answer105.0 s65.0 s

Per token is not per task: on Artificial Analysis' full suite GPT-5.6 Terra (max) cost $1.40 and Kimi K3 (max) $2.00, because Kimi K3 (max) used about 1.1× as many tokens for the same work.

Each model is compared at its best-ranked reasoning-effort setting. GPT-5.6 Terra (max) scores 64.4 and takes 101 s to its first answer token; GPT-5.6 Terra (xhigh) scores 62.8 in 7 s, GPT-5.6 Terra (high) scores 61.3 in 2 s. Kimi K3 (max) scores 64.8 and takes 53 s to its first answer token; Kimi K3 scores 58.8 in 1 s, Kimi K3 (low) scores 58.2 in 57 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

145 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.

MetricPublisherGPT-5.6 Terra (max)Kimi K3 (max)
BenchLeader IndexBenchLeader64.464.8
Agents & tools scoreBenchLeader58.959.4
Coding scoreBenchLeader65.164.6
Composite scoreBenchLeader76.678.3
Instruction following scoreBenchLeader69.9–
Knowledge scoreBenchLeader57.262.8
Long context scoreBenchLeader66.669.6
Maths scoreBenchLeader68.963.4
Multimodal scoreBenchLeader63.763.6
Reasoning scoreBenchLeader70.166.4
Human preference scoreBenchLeader–68.1
Blended price $/MBenchLeader$4.50$6.00
Output speedBenchLeader126 tok/s41 tok/s
Time to first answerBenchLeader101.0 s52.6 s
Context windowBenchLeader1,050,000 tokens1,048,576 tokens
Reasoning
GPQA DiamondEpoch AI Benchmarking Hub93.3%93.1%
SimpleBenchSimpleBench48.9%†60.7%
Arena Hard PromptsArena (LMArena)1490†1516
LiveBench 2.0 ReasoningLiveBench90.6%†90.7%†
LiveBench 1.0 ReasoningLiveBench90.6%90.7%†
GPQA Diamond (AA)Artificial Analysis92.5%93.5%
Humanity's Last Exam (AA)Artificial Analysis42.9%46.9%
GPQA Diamond (Vals)Vals AI90.9%†92.9%†
Kagi LLM BenchmarkKagi LLM Benchmark51.3%†–
ARC-AGI-1ARC Prize96.5%94.5%
ARC-AGI-2ARC Prize83.9%60.4%
ARC-AGI-3ARC Prize0.8%–
CritPtArtificial Analysis30.0%23.4%
Chess PuzzlesEpoch AI Benchmarking Hub47.0%36.0%
Mystery Game PuzzlesEpoch AI Benchmarking Hub35.0%26.0%
LMCAEpoch AI Benchmarking Hub55.0%52.7%
DTBenchEpoch AI Benchmarking Hub93.3%91.2%
ForecastBenchForecastBench–64.6%
Coding
SciCodeSciCode55.0%59.5%
WeirdMLWeirdML78.3%†82.6%
FrontierCodeCognition41.3%†44.2%†
Arena CodingArena (LMArena)1518†1539
Arena WebDevArena (LMArena)1521†1655
LiveBench 2.0 CodingLiveBench86.2%†59.3%†
LiveBench 1.0 CodingLiveBench78.3%81.5%†
Show all 145 benchmarks
MetricPublisherGPT-5.6 Terra (max)Kimi K3 (max)
Coding
LiveCodeBenchVals AI85.9%†87.2%
IOIVals AI87.6%48.9%
SWE-bench (Vals)Vals AI95.4%93.4%†
Code MigrationVals AI47.8%16.1%
Code Migration v1.1Vals AI–28.6%
ProgramBenchVals AI0.5%2.0%†
SWE-bench MultimodalVals AI–44.2%
Vibe Code Bench 1-100Vals AI14.8%18.2%
Vibe Code Bench v1.1Vals AI74.6%85.0%
Surface Evolver BenchEpoch AI Benchmarking Hub83.8%†93.0%
DeepSWE v1.1Epoch AI Benchmarking Hub69.6%68.5%
CursorBenchCursor41.3%–
ALE-BenchEpoch AI Benchmarking Hub1951.41524.5
FrontierSWEEpoch AI Benchmarking Hub–25.9%
Terminal-Bench 4.0 (AA)Artificial Analysis35.4%12.6%
Terminal-Bench 2.1 (AA)Artificial Analysis88.0%85.0%
Startrise FrontendStartrise66.9%†79.8%†
FrontierCode 1.1 (AA)Artificial Analysis–43.6%
Agents & tools
Terminal-BenchTerminal-Bench21.5%–
APEX-AgentsMercor58.2%47.8%†
Arena AgentArena (LMArena)0.2†1.6
LiveBench Long AgenticLiveBench48.4%†42.8%†
LiveBench BrowserLiveBench84.0%†16.0%†
LiveBench Agentic CodingLiveBench55.0%62.2%†
Terminal-Bench HardArtificial Analysis57.6%–
τ²-Bench Telecom (AA)Artificial Analysis86.3%–
Terminal-Bench 2.1 (Vals)Vals AI77.5%80.9%†
MCP AtlasScale AI SEAL–82.3%
HiL-BenchScale AI SEAL–42.7%†
τ³-bench Bankingτ²-bench–37.1%
GDPval-AA v2.1Artificial Analysis47.6%–
τ³-Banking (AA)Artificial Analysis40.2%46.0%
ITBench SRE (AA)Artificial Analysis51.0%47.7%
Analyst Agent (AA)Artificial Analysis–38.8%
APEX-Agents (AA)Artificial Analysis38.9%41.3%
CyberBenchVals AI72.1%75.2%
Finance Agent v2Vals AI54.4%53.1%
Finance Agent v2.1Vals AI–60.0%
Harvey's Legal Agent Benchmark v1.1Vals AI–12.9%
Harvey's Legal Agent Benchmark v1Vals AI0.8%12.9%
Legal Research BenchVals AI41.4%46.1%
SkillsBenchVals AI58.9%–
SREBenchVals AI9.5%–
Tax Agent BenchVals AI30.0%33.4%
Terminal-Bench 4.0 (Vals)Vals AI22.7%17.2%
Terminal-Bench ScienceVals AI10.0%1.4%
Time Horizon Index: KSPVals AI–10.5%
BALROGBALROG53.2%–
PostTrainBenchEpoch AI Benchmarking Hub–32.0%†
GBAEvalEpoch AI Benchmarking Hub–48.4%†
Vending-Bench 2Andon Labs7343.2†5165.0†
AutomationBenchArtificial Analysis59.6%58.3%
GDP.pdf (AA)Artificial Analysis24.0%22.0%
Harvey LAB-AA v1.1Harvey–5.3%
EnterpriseOps-GymArtificial Analysis38.5%45.3%
AA-Briefcase v1.1Artificial Analysis–1499
Terminal-Bench-Science 0.1 (AA)Artificial Analysis13.8%8.1%
CWE-Bench (AA)Artificial Analysis–53.3%
DeepSec-Bench (AA)Artificial Analysis–12.4%
CyberGym end-to-end (AA)Artificial Analysis–58.0%
FACTS SearchGoogle81.7%†–
Maths
FrontierMath Tiers 1–3Epoch AI Benchmarking Hub86.0%72.2%
FrontierMath Tier 4Epoch AI Benchmarking Hub70.7%39.0%
OTIS Mock AIMEEpoch AI Benchmarking Hub99.7%97.2%
ProofBenchVals AI74.0%†87.0%†
LiveBench MathematicsLiveBench94.9%84.4%†
AIME 2026MathArena–96.7%†
HMMT February 2026MathArena–97.0%†
MathArena ApexMathArena–65.6%†
Arena MathsArena (LMArena)1483†1499
Knowledge
SimpleQA VerifiedEpoch AI Benchmarking Hub43.2%50.6%
LiveBench 2.0 Data AnalysisLiveBench79.3%†78.7%†
LiveBench ResearchLiveBench60.4%†79.0%†
LiveBench 1.0 Data AnalysisLiveBench79.3%78.7%†
AA-OmniscienceArtificial Analysis0.119.7
MMLU-ProVals AI86.7%†88.0%
LegalBenchVals AI85.1%86.2%
CorpFinVals AI65.3%†71.6%†
TaxEvalVals AI76.2%†75.7%†
BioMysteryBenchVals AI–72.6%
Excel Modeling BenchmarkVals AI66.2%66.7%
Excel Modeling Benchmark v1.1Vals AI–71.3%
MedCodeVals AI43.4%†49.4%
MedScribeVals AI82.9%†88.0%
Public Benefits Bench v1.1Vals AI62.4%68.2%
GDP.pdfEpoch AI Benchmarking Hub24.7%†19.0%
AA-Omniscience: accuracyArtificial Analysis46.8%47.6%
AA-Omniscience: non-hallucinationArtificial Analysis12.1%46.8%
Instruction following
LiveBench LanguageLiveBench82.9%85.5%†
LiveBench Instruction FollowingLiveBench64.6%71.4%†
IFBenchArtificial Analysis71.2%–
Arena Instruction FollowingArena (LMArena)1464†1489
Human preference
Arena TextArena (LMArena)1467†1487
EQ-Bench 4EQ-Bench1224†1318†
Arena Creative WritingArena (LMArena)1429†1462
Arena Multi-turnArena (LMArena)1472†1502
Arena Longer QueriesArena (LMArena)1468†1503
Multimodal
Arena VisionArena (LMArena)1271†–
LiveBench DesignLiveBench76.0%†77.2%†
MMMU-ProArtificial Analysis80.7%80.5%
MMMU-Pro (Vals)Vals AI86.5%88.2%†
MortgageTaxVals AI67.3%†66.3%†
SAGEVals AI47.0%52.8%
Blueprint-Bench 2Epoch AI Benchmarking Hub30.8%†29.5%†
AA-3DArtificial Analysis–982
Long context
AA-LCRArtificial Analysis83.0%88.7%
Arena DocumentArena (LMArena)1477†–
MLCRArtificial Analysis27.2%21.7%
Composite
Epoch Capabilities IndexEpoch AI Benchmarking Hub160.1†157.9†
LiveBench 2.0LiveBench75.0%†63.4%†
LiveBench 1.0LiveBench77.9%79.2%†
AA Intelligence Index v4.3.2Artificial Analysis42.143.6
Vals IndexVals AI–53.0
Vals Finance IndexVals AI–58.0%
Vals RSI IndexVals AI–20.9%
Vals Multimodal IndexVals AI65.1%†73.4%†
Vals Router IndexVals AI–53.0%†
AA Cyber IndexArtificial Analysis–41.3
Safety & honesty
FORTRESSScale AI SEAL–26.6%
Metadata
AA Openness IndexArtificial Analysis–38.9

Data as of 2026-10-25. Best configuration of each model; every score links to its source on the model pages.

What changed

  • GPT-5.6 Terra (max): 1 new result (Harvey's Legal Agent Benchmark v1).
  • Kimi K3 (max): 7 new results (Code Migration v1.1, Excel Modeling Benchmark v1.1, Finance Agent v2.1…).
  • GPT-5.6 Terra (max): 1 score revision.
  • Kimi K3 (max): 1 new result (Terminal-Bench-Science 0.1 (AA)); 1 score revision.
  • Kimi K3 (max): 3 score revisions.
  • GPT-5.6 Terra (max): BenchLeader Index 63.8 → 64.4.
  • Kimi K3 (max): 1 score revision.
  • GPT-5.6 Terra (max): 5 new results (AutomationBench, GDP.pdf (AA), MLCR…).
  • Kimi K3 (max): 1 new result (τ³-bench Banking).
  • GPT-5.6 Terra (max): 1 new result (SREBench); 1 score revision.

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

GPT-5.6 Terra vs Kimi K3: questions

Is GPT-5.6 Terra better than Kimi K3?
GPT-5.6 Terra (max) and Kimi K3 (max) are neck and neck on quality (64.4 vs 64.8, within the ±6.0 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 2026-10-25, so check the category scores for your use.
Which is cheaper, GPT-5.6 Terra or Kimi K3, and will that change?
GPT-5.6 Terra is cheaper today: $4.50 against $6.00 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
Is GPT-5.6 Terra better than Kimi K3 for coding?
They are level in coding (65 vs 65 on the category index). On SciCode, the top shared coding benchmark, it is 55.0% for GPT-5.6 Terra against 59.5% for Kimi K3.
Is GPT-5.6 Terra better than Kimi K3 for agents and tool use?
They are level in agentic tasks (59 vs 59 on the category index). On τ³-Banking (AA), the top shared agents & tools benchmark, it is 40.2% for GPT-5.6 Terra against 46.0% for Kimi K3.
Which is better for long documents?
Kimi K3 scores higher on long-context benchmarks (70 vs 67). Both take about 1.1M tokens: GPT-5.6 Terra 1,050,000, Kimi K3 1,048,576.
Which has lower latency?
Kimi K3 (max) answers first: 52.6 s against 101.0 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.
Which is faster, GPT-5.6 Terra or Kimi K3?
GPT-5.6 Terra streams faster: 126 against 41 output tokens per second.
Do independent benchmarks agree on GPT-5.6 Terra vs Kimi K3?
Across the 47 benchmarks both have, GPT-5.6 Terra leads 26, Kimi K3 leads 21. Every overall index that rates both puts Kimi K3 ahead.
When was Kimi K3 released?
Kimi K3 was released on 16 Jul 2026.