Agentic
Directional only- Claude Opus 5
- 77.4
- Supported · #2/151
- Kimi K2.5
- 46.0
- Estimated · #90/151
- Basis
- BenchAlign lane · 19 vs 14 public rows
- Reading
- Directional only
Claude Opus 5 vs Kimi K2.5
Updated September 4, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.
Decision reading
Claude Opus 5 has the higher public score, 80.66 versus 55.62, and the 90% score intervals do not overlap.
7 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.
Share or export
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.
Long documents
Prompts that approach the documented context limit
Claude Opus 5
Claude Opus 5 has the larger documented context window.
Confidence: documented
Chat turn cost
1K fresh input + 500 output tokens
Kimi K2.5
Kimi K2.5 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Cache-heavy agent loop cost
200K cached + 20K fresh input + 10K output tokens
Kimi K2.5
Kimi K2.5 has the lower estimated token cost for this stated workload. Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate.
Confidence: rate-fallback
Show secondary and unsupported calls
Repository review cost
50K fresh input + 3K output tokens
Kimi K2.5
Kimi K2.5 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Coding work
Code generation, repair, and software-engineering tasks
Not enough matched evidence
Kimi K2.5 is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like.
Confidence: limited
Agentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
Kimi K2.5 is scored on Estimated evidence for agentic, so the reading is directional rather than like-for-like.
Confidence: limited
What is actually comparable
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
66 Claude Opus 5 only
7 Shared
38 Kimi K2.5 only
- Shared results
- 7
- Claude Opus 5 only
- 66
- Kimi K2.5 only
- 38
- Like-for-like categories
- 0 / 8
4 categories rest on Estimated evidence or different evidence sets. Those rows remain visible for coverage context but do not name a winner.
Category results, on a stated basis
Each row shows the public-lane category score for both models: the BenchAlign lane for agentic, coding, and knowledge, and the weighted public rows elsewhere, with the same rank each model page reports. A row is like-for-like only when both scores rest on Supported evidence or the same weighted set. Directional and not-comparable rows remain visible, but they never receive a winner.
Coding
Directional only- Claude Opus 5
- 75.6
- Supported · #3/183
- Kimi K2.5
- 51.5
- Estimated · #64/183
- Basis
- BenchAlign lane · 16 vs 8 public rows
- Reading
- Directional only
Knowledge
Directional only- Claude Opus 5
- 82.1
- Supported · #3/181
- Kimi K2.5
- 53.1
- Estimated · #70/181
- Basis
- BenchAlign lane · 19 vs 6 public rows
- Reading
- Directional only
Multimodal
Directional only- Claude Opus 5
- 88.7
- #2/48
- Kimi K2.5
- 65.7
- #24/48
- Basis
- Provisional lane · 1 vs 1 weighted rows
- Reading
- Directional only
Reasoning
Not comparable- Claude Opus 5
- 75.6
- #13/22
- Kimi K2.5
- 53.5
- Unranked · 3 rankable rows
- Basis
- Provisional lane · 2 vs 1 weighted rows
- Reading
- Not comparable
Math
Not comparable- Claude Opus 5
- Not ranked
- Kimi K2.5
- 62.4
- #5/7
- Basis
- Provisional lane · 0 vs 4 weighted rows
- Reading
- Not comparable
Multilingual
Not comparable- Claude Opus 5
- Not ranked
- Kimi K2.5
- 38.2
- #8/12
- Basis
- Provisional lane · 0 vs 1 weighted rows
- Reading
- Not comparable
Instruction following
Not comparable- Claude Opus 5
- Not ranked
- Kimi K2.5
- 85.6
- #40/120
- Basis
- Provisional lane · 0 vs 0 weighted rows
- Reading
- Not comparable
| Category | Claude Opus 5 | Kimi K2.5 | Basis | Reading |
|---|---|---|---|---|
| Agentic | 77.4Supported · #2/151 | 46.0Estimated · #90/151 | Directional onlyBenchAlign lane · 19 vs 14 public rows | Directional only |
| Coding | 75.6Supported · #3/183 | 51.5Estimated · #64/183 | Directional onlyBenchAlign lane · 16 vs 8 public rows | Directional only |
| Knowledge | 82.1Supported · #3/181 | 53.1Estimated · #70/181 | Directional onlyBenchAlign lane · 19 vs 6 public rows | Directional only |
| Multimodal | 88.7#2/48 | 65.7#24/48 | Directional onlyProvisional lane · 1 vs 1 weighted rows | Directional only |
| Reasoning | 75.6#13/22 | 53.5Unranked · 3 rankable rows | Not comparableProvisional lane · 2 vs 1 weighted rows | Not comparable |
| Math | Not ranked | 62.4#5/7 | Not comparableProvisional lane · 0 vs 4 weighted rows | Not comparable |
| Multilingual | Not ranked | 38.2#8/12 | Not comparableProvisional lane · 0 vs 1 weighted rows | Not comparable |
| Instruction following | Not ranked | 85.6#40/120 | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
Ranks count the models scored in each category’s lane, so the agentic, coding, and knowledge denominators (BenchAlign) differ from the provisional-lane categories. Unranked scores sit on the lane’s scale but never name a winner. Like-for-like rows within 0.5 points read as a practical tie.
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.
HLE
Knowledge
BrowseComp
Agentic
SWE-bench Pro
Coding
SWE-bench Verified
Coding
SWE Multilingual
Coding
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
- Claude Opus 5
- $0.0175
- Fits in one request
- Kimi K2.5
- $0.0021
- Fits in one request
Kimi K2.5 has the lower modeled cost
Costs use the listed standard API rates.
Repository review
50K fresh input + 3K output tokens
- Claude Opus 5
- $0.325
- Fits in one request
- Kimi K2.5
- $0.039
- Fits in one request
Kimi K2.5 has the lower modeled cost
Costs use the listed standard API rates.
Cache-heavy agent loop
200K cached + 20K fresh input + 10K output tokens
- Claude Opus 5
- $0.45
- Fits in one request
- Kimi K2.5
- $0.162
- Fits in one request
- Cached input priced at the published list-input rate
Kimi K2.5 has the lower modeled cost
Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate.
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationClaude Opus 5Kimi K2.5
Context window
Maximum documented context; output-token limits may be lower.
Claude Opus 5
Kimi K2.5
256K
API model ID
Claude Opus 5
claude-opus-5
Anthropic model overviewKimi 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.
Claude Opus 5
$0.5 per 1M cached input tokens
Claude API pricingKimi K2.5
Not published
Documented inputs
Claude Opus 5
text, image
Anthropic model overviewKimi K2.5
Not sourced
Documented outputs
Claude Opus 5
Kimi K2.5
Not sourced
Provider availability
Claude Opus 5
Generally Available · Claude API
Anthropic model overviewKimi K2.5
Not sourced
Reasoning profile
Claude Opus 5
Reasoning
Kimi K2.5
Non-Reasoning
Weight access
Claude Opus 5
Proprietary
Kimi K2.5
Open Weight
License
Claude Opus 5
Proprietary
Kimi K2.5
Open Weight
Release date
Claude Opus 5
2026-07-24
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
- Claude Opus 5 has the higher public score, 80.66 versus 55.62, and the 90% score intervals do not overlap.
- Workload cost
- Repository review: $0.325 vs $0.039. Cache-heavy agent loop: $0.45 vs $0.162.
- Context tradeoff
- Claude Opus 5 has the larger documented window (1M).
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.
Claude Opus 5
API / mo$22,500
Self-host / moNot listed
Break-even—
Proprietary model — self-hosting not applicable.
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 evidence111 rows
Agentic
Terminal-Bench 3.0
Claude Opus 542.7%SourceKimi K2.5—Not directly comparable
BrowseComp
Claude Opus 590.8%SourceKimi K2.560.6%SourceClaude Opus 5 leads this result
HLE w/ tools
Claude Opus 564.7%SourceKimi K2.5—Not directly comparable
DeepSearchQA
Claude Opus 595.0%SourceKimi K2.577.1%SourceClaude Opus 5 leads this result
DRACO
Claude Opus 588.6%SourceKimi K2.5—Not directly comparable
BrowseComp (10-agent, prerelease)
Claude Opus 593.6%SourceKimi K2.5—Not directly comparable
OSWorld 2.0
Claude Opus 570.6%SourceKimi K2.5—Not directly comparable
MCP Atlas
Claude Opus 585.8%SourceKimi K2.529.5%SourceClaude Opus 5 leads this result
MCP-Atlas claim coverage
Claude Opus 589.1%SourceKimi K2.5—Not directly comparable
LAB all-pass (Anthropic harness)
Claude Opus 523.58%SourceKimi K2.5—Not directly comparable
LAB criterion-pass (Anthropic harness)
Claude Opus 593.74%SourceKimi K2.5—Not directly comparable
LAB all-pass (Harvey held-out)
Claude Opus 511.7%SourceKimi K2.5—Not directly comparable
LAB criterion-pass (Harvey held-out)
Claude Opus 594.1%SourceKimi K2.5—Not directly comparable
Toolathlon-Verified
Claude Opus 580.6%SourceKimi K2.5—Not directly comparable
Toolathlon Verified Pass@3
Claude Opus 587.0%SourceKimi K2.5—Not directly comparable
Toolathlon Verified Pass³
Claude Opus 573.1%SourceKimi K2.5—Not directly comparable
Toolathlon Verified avg. turns
Claude Opus 523.5 turnsSourceKimi K2.5—Not directly comparable
AutomationBench
Claude Opus 526.0%SourceKimi K2.5—Not directly comparable
Terminal-Bench 2.1 (Vals)
Claude Opus 584.6%SourceKimi K2.5—Not directly comparable
Terminal-Bench 2.0
Claude Opus 5—Kimi K2.550.8%SourceNot directly comparable
Claw-Eval
Claude Opus 5—Kimi K2.552.3%SourceNot directly comparable
QwenClawBench
Claude Opus 5—Kimi K2.554.3%SourceNot directly comparable
τ³-bench results
Claude Opus 5—Kimi K2.565.7%SourceNot directly comparable
DeepPlanning
Claude Opus 5—Kimi K2.514.4%SourceNot directly comparable
Toolathlon
Claude Opus 5—Kimi K2.527.8%SourceNot directly comparable
MCP-Tasks
Claude Opus 5—Kimi K2.559.1%SourceNot directly comparable
WideResearch
Claude Opus 5—Kimi K2.572.7%SourceNot directly comparable
Gert Labs
Claude Opus 5—Kimi K2.545.88%SourceNot directly comparable
ResearchClawBench
Claude Opus 5—Kimi K2.514.0%SourceNot directly comparable
JobBench
Claude Opus 5—Kimi K2.58.7%SourceNot directly comparable
Coding
Bug Hunt Bench
Claude Opus 527 fixesSourceKimi K2.5—Not directly comparable
SWE-bench Verified
Claude Opus 596%SourceKimi K2.576.8%SourceClaude Opus 5 leads this result
SWE-bench Pro
Claude Opus 579.2%SourceKimi K2.550.7%SourceClaude Opus 5 leads this result
SWE Multilingual
Claude Opus 589.5%SourceKimi K2.573%SourceClaude Opus 5 leads this result
SWE Multimodal
Claude Opus 559.4%SourceKimi K2.5—Not directly comparable
deepSwe
Claude Opus 568.8%SourceKimi K2.5—Not directly comparable
FrontierCode 1.1 Main
Claude Opus 553.4%SourceKimi K2.5—Not directly comparable
FrontierCode 1.1 Extended
Claude Opus 563.6%SourceKimi K2.5—Not directly comparable
FrontierSWE v2
Claude Opus 552.0%SourceKimi K2.5—Not directly comparable
ProgramBench (episode 1)
Claude Opus 583.0%SourceKimi K2.5—Not directly comparable
ProgramBench
Claude Opus 593.0%SourceKimi K2.5—Not directly comparable
cursorBench32
Claude Opus 570.0%SourceKimi K2.5—Not directly comparable
VulcanBench v3
Claude Opus 587.0%SourceKimi K2.5—Not directly comparable
VulcanBench CII v1
Claude Opus 596.4%SourceKimi K2.5—Not directly comparable
LiveCodeBench (Vals)
Claude Opus 589.0%SourceKimi K2.5—Not directly comparable
SWE-bench (Vals)
Claude Opus 597.0%SourceKimi K2.5—Not directly comparable
SWE-bench Verified*
Claude Opus 5—Kimi K2.570.8%SourceNot directly comparable
LiveCodeBench v6
Claude Opus 5—Kimi K2.585.0%SourceNot directly comparable
SWE-Rebench
Claude Opus 5—Kimi K2.558.5%SourceNot directly comparable
React Native Evals
Claude Opus 5—Kimi K2.577.2%SourceNot directly comparable
SciCode
Claude Opus 5—Kimi K2.548.7%SourceNot directly comparable
Reasoning
Knowledge
HLE
Claude Opus 564.7%SourceKimi K2.530.1%SourceClaude Opus 5 leads this result
HLE w/o tools
Claude Opus 556.3%SourceKimi K2.5—Not directly comparable
HLE-Verified
Claude Opus 554.4%SourceKimi K2.5—Not directly comparable
LABBench2
Claude Opus 584.2%SourceKimi K2.5—Not directly comparable
HealthBench (raw)
Claude Opus 567.1%SourceKimi K2.5—Not directly comparable
HealthBench (length-adjusted)
Claude Opus 557.8%SourceKimi K2.5—Not directly comparable
HealthBench Professional
Claude Opus 559.8%SourceKimi K2.5—Not directly comparable
HealthBench Professional (raw)
Claude Opus 573.4%SourceKimi K2.5—Not directly comparable
BioMysteryBench (human-solvable)
Claude Opus 590.1%SourceKimi K2.5—Not directly comparable
BioMysteryBench (human-difficult)
Claude Opus 549.4%SourceKimi K2.5—Not directly comparable
SpatialBench Verified
Claude Opus 572.5%SourceKimi K2.5—Not directly comparable
SingleCellBench
Claude Opus 560.6%SourceKimi K2.5—Not directly comparable
ProteinGym Hard
Claude Opus 547.7%SourceKimi K2.5—Not directly comparable
Protein Design
Claude Opus 542.5%SourceKimi K2.5—Not directly comparable
Organic chemistry V2
Claude Opus 561.6%SourceKimi K2.5—Not directly comparable
Protocols (troubleshooting)
Claude Opus 561.1%SourceKimi K2.5—Not directly comparable
Protocols (understanding)
Claude Opus 578.4%SourceKimi K2.5—Not directly comparable
GPQA Diamond (Vals)
Claude Opus 593.4%SourceKimi K2.5—Not directly comparable
MMLU-Pro (Vals)
Claude Opus 591.6%SourceKimi K2.5—Not directly comparable
GPQA
Claude Opus 5—Kimi K2.587.6%SourceNot directly comparable
GPQA-D
Claude Opus 5—Kimi K2.587.6%SourceNot directly comparable
SuperGPQA
Claude Opus 5—Kimi K2.569.2%SourceNot directly comparable
MMLU-Pro
Claude Opus 5—Kimi K2.587.1%SourceNot directly comparable
MMLU-Pro (Arcee)
Claude Opus 5—Kimi K2.587.1%SourceNot directly comparable
Math
IMO 2026
Claude Opus 542/42SourceKimi K2.5—Not directly comparable
RiemannBench (no tools)
Claude Opus 560.0%SourceKimi K2.5—Not directly comparable
RiemannBench (tools)
Claude Opus 579.0%SourceKimi K2.5—Not directly comparable
ArXivMath Jun. 2026 (no tools)
Claude Opus 590.8%SourceKimi K2.5—Not directly comparable
ArXivMath Jun. 2026 (tools)
Claude Opus 591.3%SourceKimi K2.5—Not directly comparable
AIME 2025
Claude Opus 5—Kimi K2.596.1%SourceNot directly comparable
AIME26
Claude Opus 5—Kimi K2.595.8%SourceNot directly comparable
AIME25 (Arcee)
Claude Opus 5—Kimi K2.596.3%SourceNot directly comparable
HMMT Feb 2025
Claude Opus 5—Kimi K2.595.4%SourceNot directly comparable
HMMT Nov 2025
Claude Opus 5—Kimi K2.591.1%SourceNot directly comparable
HMMT Feb 2026
Claude Opus 5—Kimi K2.587.1%SourceNot directly comparable
MMAnswerBench
Claude Opus 5—Kimi K2.581.8%SourceNot directly comparable
FrontierMath v2 (Tiers 1-3)
Claude Opus 5—Kimi K2.527.900%SourceNot directly comparable
FrontierMath v2 (Tier 4)
Claude Opus 5—Kimi K2.54.200%SourceNot directly comparable
Multilingual
GMMLU
Claude Opus 592.5%SourceKimi K2.5—Not directly comparable
MILU
Claude Opus 592.1%SourceKimi K2.5—Not directly comparable
INCLUDE
Claude Opus 589.8%SourceKimi K2.5—Not directly comparable
MMLU-ProX
Claude Opus 5—Kimi K2.582.3%SourceNot directly comparable
NOVA-63
Claude Opus 5—Kimi K2.556.0%SourceNot directly comparable
Multimodal
Chartography (no tools)
Claude Opus 529.6%SourceKimi K2.5—Not directly comparable
Chartography (tools)
Claude Opus 583.0%SourceKimi K2.5—Not directly comparable
BenchCAD Vision2Code (no tools)
Claude Opus 50.366SourceKimi K2.5—Not directly comparable
BenchCAD Vision2Code (tools)
Claude Opus 50.821SourceKimi K2.5—Not directly comparable
GDP.pdf (no tools)
Claude Opus 583.4%SourceKimi K2.5—Not directly comparable
GDP.pdf (tools)
Claude Opus 585.5%SourceKimi K2.5—Not directly comparable
OfficeQA
Claude Opus 578.1%SourceKimi K2.5—Not directly comparable
OfficeQA Pro
Claude Opus 566.9%SourceKimi K2.5—Not directly comparable
MMMU-Pro
Claude Opus 5—Kimi K2.578.5%SourceNot directly comparable
Video-MME
Claude Opus 5—Kimi K2.587.4%SourceNot directly comparable
MMVU
Claude Opus 5—Kimi K2.580.4%SourceNot directly comparable
VideoMMMU
Claude Opus 5—Kimi K2.586.6%SourceNot directly comparable
Instruction following
IFEval
Claude Opus 5—Kimi K2.593.9%SourceNot directly comparable
Frequently asked questions
Which is better, Claude Opus 5 or Kimi K2.5?
Claude Opus 5 has the higher public score, 80.66 versus 55.62, and the 90% score intervals do not overlap. That is the clearest overall quality signal in the current public data.
Which is better for coding, Claude Opus 5 or Kimi K2.5?
Claude Opus 5 scores higher for coding on the public lane, 75.6 to 51.5. Kimi K2.5 is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like. BenchLM does not name a winner for coding from a directional reading; read the shared benchmark rows directly and test the models on the same task set.
Which is better for agentic tasks, Claude Opus 5 or Kimi K2.5?
Claude Opus 5 scores higher for agentic tasks on the public lane, 77.4 to 46. Kimi K2.5 is scored on Estimated evidence for agentic tasks, so the reading is directional rather than like-for-like. BenchLM does not name a winner for agentic tasks from a directional reading; read the shared benchmark rows directly and test the models on the same task set.
Which costs less, Claude Opus 5 or Kimi K2.5?
For the stated presets, chat costs $0.0175 on Claude Opus 5 and $0.0021 on Kimi K2.5; repository review costs $0.325 and $0.039; the cache-heavy agent loop costs $0.45 and $0.162. Kimi K2.5 has no published cached-input rate, so cached tokens use its listed input rate.
Which has the larger context window, Claude Opus 5 or Kimi K2.5?
Claude Opus 5 has the larger documented context window: 1M, compared with 256K.
Related comparisons
Last updated September 4, 2026
Watch Claude Opus 5 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.