Model comparison
GPT-5.2 vs Kimi K2.5
Head-to-head evidence from 21 shared benchmark results across 7 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.
Public leaderboard positions: GPT-5.2 #64 (Estimated); Kimi K2.5 #54 (Supported). Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.
Evidence parity. GPT-5.2 and Kimi K2.5 share 21 comparable benchmark results. 6 of 8 categories are comparable. 7 results are unique to GPT-5.2; 42 to Kimi K2.5.
Updated July 20, 2026- Shared results
- 21
- GPT-5.2 only
- 7
- Kimi K2.5 only
- 42
- Comparable categories
- 6 / 8
Pick Kimi K2.5 if you want the stronger benchmark profile. GPT-5.2 only becomes the better choice if knowledge is the priority or you need the larger 400K context window.
Confidence note. This is a partial-evidence comparison with 21 shared benchmark results across 7 evidence categories; 6 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.
Why this result
Kimi K2.5 has the cleaner BenchAlign overall profile here, landing at 59.66 versus 58.43. It is a real lead, but still close enough that category-level strengths matter more than the headline number.
Kimi K2.5's sharpest advantage is in mathematics, where it averages 60.6 against 35.2. The single biggest benchmark swing on the page is FrontierMath v2 (Tier 4), 18.800% to 4.200%. GPT-5.2 does hit back in knowledge, so the answer changes if that is the part of the workload you care about most.
GPT-5.2 is also the more expensive model on tokens at $1.75 input / $14.00 output per 1M tokens, versus $0.60 input / $3.00 output per 1M tokens for Kimi K2.5. That is roughly 4.7x on output cost alone. GPT-5.2 is the reasoning model in the pair, while Kimi K2.5 is not. That usually helps on harder chain-of-thought-heavy tests, but it can also mean more latency and more token spend in real use. GPT-5.2 gives you the larger context window at 400K, compared with 256K for Kimi K2.5.
Category breakdown
Exact category averages are shown below. Not measured means BenchLM does not have enough sourced public coverage for that model and category.
| Category | GPT-5.2 | Δ | Kimi K2.5 |
|---|---|---|---|
| Knowledge | GPT-5.292.4 | Margin← 35.5 | Kimi K2.556.9 |
| Math | GPT-5.235.2 | Margin→ 25.4 | Kimi K2.560.6 |
| Coding | GPT-5.270.6 | Margin← 11.2 | Kimi K2.559.4 |
| Reasoning | GPT-5.252.9 | Margin→ 8.1 | Kimi K2.561.0 |
| Multimodal | GPT-5.280.4 | Margin← 1.9 | Kimi K2.578.5 |
| Agentic | GPT-5.255.7 | Margin← 0.7 | Kimi K2.555.0 |
| Multilingual | GPT-5.2Not measured | MarginNo overlap | Kimi K2.582.3 |
| Inst. Following | GPT-5.2Not measured | MarginNo overlap | Kimi K2.593.9 |
Decisive benchmark drivers
The largest measured benchmark gaps in this matchup, with exact reported values.
More
- Source ↗
FrontierMath v2 (Tier 4)
MathA 18.800%B 4.200%Winner: GPT-5.2Δ 14.6FrontierMath v2 (Tier 4): GPT-5.2 scored 18.800%; Kimi K2.5 scored 4.200%. GPT-5.2 wins this benchmark. - Source ↗
FrontierMath v2 (Tiers 1-3)
MathA 40.700%B 27.900%Winner: GPT-5.2Δ 12.8FrontierMath v2 (Tiers 1-3): GPT-5.2 scored 40.700%; Kimi K2.5 scored 27.900%. GPT-5.2 wins this benchmark. - Source ↗
BrowseComp
AgenticA 65.8%B 60.6%Winner: GPT-5.2Δ 5.2BrowseComp: GPT-5.2 scored 65.8%; Kimi K2.5 scored 60.6%. GPT-5.2 wins this benchmark. - Source ↗
SWE-bench Pro
CodingA 55.6%B 50.7%Winner: GPT-5.2Δ 4.9SWE-bench Pro: GPT-5.2 scored 55.6%; Kimi K2.5 scored 50.7%. GPT-5.2 wins this benchmark. - Source ↗
GPQA
KnowledgeA 92.4%B 87.6%Winner: GPT-5.2Δ 4.8GPQA: GPT-5.2 scored 92.4%; Kimi K2.5 scored 87.6%. GPT-5.2 wins this benchmark.
Operational comparison
Runtime and commercial metrics are compared only when both models have a complete sourced value.
| Metric | GPT-5.2 | Kimi K2.5 | Comparison |
|---|---|---|---|
| Input / output priceUSD per 1M tokens | GPT-5.2$1.75 input / $14 output | Kimi K2.5$0.6 input / $3 output | Kimi K2.5 has the lower combined listed price. |
| Generation speedtokens per second | GPT-5.273 tok/s | Kimi K2.545 tok/s | GPT-5.2 has the higher measured throughput. |
| First-answer latencyseconds to first token | GPT-5.2130.34 s | Kimi K2.52.38 s | Kimi K2.5 reaches the first token sooner. |
| Context windowmaximum listed tokens | GPT-5.2400K | Kimi K2.5256K | GPT-5.2 lists the larger context window. |
Benchmark Deep Dive
AgenticGPT-5.2 wins20 benchmarks
| Benchmark | GPT-5.2 | Kimi K2.5 | Result |
|---|---|---|---|
| BrowseCompSource | 65.8% | 60.6% | GPT-5.2 leads |
| OSWorld-VerifiedSource | 47.3% | — | Not comparable |
| τ²-bench resultsSource | 84.8% | 95.9% | Kimi K2.5 leads |
| Gert LabsSource | 46.54% | 45.88% | GPT-5.2 leads |
| JobBenchSource | 34.3% | 8.7% | GPT-5.2 leads |
| Terminal-Bench 2.0Source | — | 50.8% | Not comparable |
| Claw-EvalSource | — | 52.3% | Not comparable |
| QwenClawBenchSource | — | 54.3% | Not comparable |
| τ³-bench resultsSource | — | 65.7% | Not comparable |
| DeepSearchQASource | — | 77.1% | Not comparable |
| DeepPlanningSource | — | 14.4% | Not comparable |
| ToolathlonSource | — | 27.8% | Not comparable |
| MCP AtlasSource | — | 29.5% | Not comparable |
| MCP-TasksSource | — | 59.1% | Not comparable |
| WideResearchSource | — | 72.7% | Not comparable |
| APEX-Agents-AASource | — | 11.5% | Not comparable |
| ResearchClawBenchSource | — | 14.0% | Not comparable |
| AA Agentic IndexSource | — | 21.7% | Not comparable |
| GDPval-AASource | — | 25.4% | Not comparable |
| GDPval-AASource | — | 1009 | Not comparable |
CodingGPT-5.2 wins11 benchmarks
| Benchmark | GPT-5.2 | Kimi K2.5 | Result |
|---|---|---|---|
| SWE-bench VerifiedSource | 80% | 76.8% | GPT-5.2 leads |
| SWE-bench ProSource | 55.6% | 50.7% | GPT-5.2 leads |
| Vibe Code BenchSource | 53.50% | — | Not comparable |
| AA-SciCodeSource | 52.1% | 49.0% | GPT-5.2 leads |
| SWE-bench Verified*Source | — | 70.8% | Not comparable |
| LiveCodeBench v6Source | — | 85.0% | Not comparable |
| SWE MultilingualSource | — | 73% | Not comparable |
| SWE-RebenchSource | — | 58.5% | Not comparable |
| React Native EvalsSource | — | 77.2% | Not comparable |
| SciCodeSource | — | 48.7% | Not comparable |
| AA Coding IndexSource | — | 46.8% | Not comparable |
ReasoningKimi K2.5 wins4 benchmarks
KnowledgeGPT-5.2 wins12 benchmarks
| Benchmark | GPT-5.2 | Kimi K2.5 | Result |
|---|---|---|---|
| GPQASource | 92.4% | 87.6% | GPT-5.2 leads |
| Artificial Analysis Intelligence IndexSource | 42.2% | 35.4% | GPT-5.2 leads |
| AA-GPQA DiamondSource | 90.3% | 87.9% | GPT-5.2 leads |
| AA-HLESource | 35.4% | 29.4% | GPT-5.2 leads |
| AA-Omniscience IndexSource | -1.0% | -8.1% | GPT-5.2 leads |
| AA-Omniscience AccuracySource | 43.8% | 34.3% | GPT-5.2 leads |
| AA-Omniscience Hallucination RateSource | 79.7% | 64.6% | Kimi K2.5 leads |
| GPQA-DSource | — | 87.6% | Not comparable |
| SuperGPQASource | — | 69.2% | Not comparable |
| MMLU-ProSource | — | 87.1% | Not comparable |
| MMLU-Pro (Arcee)Source | — | 87.1% | Not comparable |
| HLESource | — | 30.1% | Not comparable |
MathKimi K2.5 wins10 benchmarks
| Benchmark | GPT-5.2 | Kimi K2.5 | Result |
|---|---|---|---|
| AA AIME 2025Source | 99.0% | — | Not comparable |
| FrontierMath v2 (Tiers 1-3)Source | 40.700% | 27.900% | GPT-5.2 leads |
| FrontierMath v2 (Tier 4)Source | 18.800% | 4.200% | GPT-5.2 leads |
| AIME 2025Source | — | 96.1% | Not comparable |
| AIME26Source | — | 95.8% | Not comparable |
| AIME25 (Arcee)Source | — | 96.3% | Not comparable |
| HMMT Feb 2025Source | — | 95.4% | Not comparable |
| HMMT Nov 2025Source | — | 91.1% | Not comparable |
| HMMT Feb 2026Source | — | 87.1% | Not comparable |
| MMAnswerBenchSource | — | 81.8% | Not comparable |
Multilingual2 benchmarks
MultimodalGPT-5.2 wins9 benchmarks
| Benchmark | GPT-5.2 | Kimi K2.5 | Result |
|---|---|---|---|
| MMMU-ProSource | 79.5% | 78.5% | GPT-5.2 leads |
| MathVisionSource | 83.0% | — | Not comparable |
| CharXivSource | 82.1% | — | Not comparable |
| V*Source | 75.9% | — | Not comparable |
| Design Arena WebsiteSource | 1227 | 1282 | Kimi K2.5 leads |
| Video-MMESource | — | 87.4% | Not comparable |
| MMVUSource | — | 80.4% | Not comparable |
| VideoMMMUSource | — | 86.6% | Not comparable |
| AA-MMMU-ProSource | — | 75.4% | Not comparable |
Frequently Asked Questions (7)
Which is better, GPT-5.2 or Kimi K2.5?
Kimi K2.5 is ahead on BenchLM's BenchAlign leaderboard, 59.66 to 58.43. The biggest single separator in this matchup is FrontierMath v2 (Tier 4), where the scores are 18.800% and 4.200%.
Which is better for knowledge tasks, GPT-5.2 or Kimi K2.5?
GPT-5.2 has the edge for knowledge tasks in this comparison, averaging 92.4 versus 56.9. Inside this category, AA-Omniscience Hallucination Rate is the benchmark that creates the most daylight between them.
Which is better for coding, GPT-5.2 or Kimi K2.5?
GPT-5.2 has the edge for coding in this comparison, averaging 70.6 versus 59.4. Inside this category, SWE-bench Pro is the benchmark that creates the most daylight between them.
Which is better for math, GPT-5.2 or Kimi K2.5?
Kimi K2.5 has the edge for math in this comparison, averaging 60.6 versus 35.2. Inside this category, FrontierMath v2 (Tier 4) is the benchmark that creates the most daylight between them.
Which is better for reasoning, GPT-5.2 or Kimi K2.5?
Kimi K2.5 has the edge for reasoning in this comparison, averaging 61 versus 52.9. Inside this category, CritPt is the benchmark that creates the most daylight between them.
Which is better for agentic tasks, GPT-5.2 or Kimi K2.5?
GPT-5.2 has the edge for agentic tasks in this comparison, averaging 55.7 versus 55. Inside this category, JobBench is the benchmark that creates the most daylight between them.
Which is better for multimodal and grounded tasks, GPT-5.2 or Kimi K2.5?
GPT-5.2 has the edge for multimodal and grounded tasks in this comparison, averaging 80.4 versus 78.5. Inside this category, Design Arena Website is the benchmark that creates the most daylight between them.
Self-host vs API cost
Estimates at 50,000 req/day · 1000 tokens/req average.
Related Comparisons
Explore More
Choose a model with this week’s evidence
Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.
One email each week. Unsubscribe anytime.