Agentic
Like-for-like- Claude Opus 5
- 78.1
- Supported · #2/153
- Kimi K3
- 72.0
- Supported · #4/153
- Basis
- BenchAlign lane · 20 vs 12 public rows
- Reading
- Claude Opus 5 leads · intervals overlap
Keep up with the models you depend on. Follow price changes, retirements, and API updates.Follow the models you depend on.
Follow model changesUpdated September 14, 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 estimate, 81.89 versus 74.9, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.
19 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
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.
Code generation, repair, and software-engineering tasks
Claude Opus 5
Claude Opus 5 leads on the public coding lane, 75.9 to 68, with Supported evidence for both models, although the 90% intervals overlap.
Confidence: limited
Tool use, computer use, and multi-step task completion
Claude Opus 5
Claude Opus 5 leads on the public agentic lane, 78.1 to 72, with Supported evidence for both models, although the 90% intervals overlap.
Confidence: limited
Prompts that approach the documented context limit
Kimi K3
Kimi K3 has the larger documented context window.
Confidence: documented
1K fresh input + 500 output tokens
Kimi K3
Kimi K3 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
200K cached + 20K fresh input + 10K output tokens
Kimi K3
Kimi K3 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
50K fresh input + 3K output tokens
Kimi K3
Kimi K3 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
2 categories rest on Estimated evidence or different evidence sets. Those rows remain visible for coverage context but do not name a winner.
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.
| Category | Claude Opus 5 | Kimi K3 | Basis | Reading |
|---|---|---|---|---|
| Agentic | 78.1Supported · #2/153 | 72.0Supported · #4/153 | Like-for-likeBenchAlign lane · 20 vs 12 public rows | Claude Opus 5 leads · intervals overlap |
| Coding | 75.9Supported · #3/152 | 68.0Supported · #6/152 | Like-for-likeBenchAlign lane · 17 vs 13 public rows | Claude Opus 5 leads · intervals overlap |
| Knowledge | 82.2Supported · #3/183 | 72.0Supported · #8/183 | Like-for-likeBenchAlign lane · 19 vs 6 public rows | Claude Opus 5 leads · intervals overlap |
| Reasoning | 75.7#11/20 | 78.5#3/20 | Directional onlyProvisional lane · 2 vs 0 weighted rows | Directional only |
| Multimodal | 88.7#2/48 | 89.5#1/48 | Directional onlyProvisional lane · 1 vs 3 weighted rows | Directional only |
| Math | Not ranked | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| Multilingual | Not ranked | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| Instruction following | Not ranked | Not ranked | 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.
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 w/o tools
Knowledge
cursorBench32
Coding
HLE
Knowledge
AutomationBench
Agentic
OfficeQA Pro
Multimodal
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.
1K fresh input + 500 output tokens
Kimi K3 has the lower modeled cost
Costs use the listed standard API rates.
50K fresh input + 3K output tokens
Kimi K3 has the lower modeled cost
Costs use the listed standard API rates.
200K cached + 20K fresh input + 10K output tokens
Kimi K3 has the lower modeled cost
Costs use the listed standard API rates.
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
Maximum documented context; output-token limits may be lower.
Claude Opus 5
Kimi K3
1.05M
Claude Opus 5
claude-opus-5
Anthropic model overviewKimi K3
Not sourced
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 K3
$0.3 per 1M cached input tokens
Claude Opus 5
text, image
Anthropic model overviewKimi K3
Not sourced
Claude Opus 5
Kimi K3
Not sourced
Claude Opus 5
Generally Available · Claude API
Anthropic model overviewKimi K3
Not sourced
Claude Opus 5
Reasoning
Kimi K3
Reasoning
Claude Opus 5
Proprietary
Kimi K3
Pending
Claude Opus 5
Proprietary
Kimi K3
Pending
Claude Opus 5
2026-07-24
Kimi K3
2026-07-16
Run the same representative tasks against both endpoints before changing production traffic.
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Terminal-Bench 3.0
Not directly comparable
BrowseComp
Kimi K3 leads this result
HLE w/ tools
Not directly comparable
DeepSearchQA
Tie
DRACO
Not directly comparable
BrowseComp (10-agent, prerelease)
Not directly comparable
OSWorld 2.0
Not directly comparable
MCP Atlas
Claude Opus 5 leads this result
MCP-Atlas claim coverage
Not directly comparable
LAB all-pass (Anthropic harness)
Not directly comparable
LAB criterion-pass (Anthropic harness)
Not directly comparable
LAB all-pass (Harvey held-out)
Not directly comparable
LAB criterion-pass (Harvey held-out)
Not directly comparable
Toolathlon-Verified
Claude Opus 5 leads this result
Toolathlon Verified Pass@3
Not directly comparable
Toolathlon Verified Pass³
Not directly comparable
Toolathlon Verified avg. turns
Not directly comparable
AutomationBench
Kimi K3 leads this result
Terminal-Bench 2.1 (Vals)
Claude Opus 5 leads this result
ApprenticeBench
Shared sourceClaude Opus 5 leads this result
Terminal-Bench 2.0
Not directly comparable
JobBench
Not directly comparable
APEX-Agents
Not directly comparable
SpreadsheetBench 2
Not directly comparable
DECK-Bench
Not directly comparable
Bug Hunt Bench
Not directly comparable
SWE-bench Verified
Not directly comparable
SWE-bench Pro
Not directly comparable
SWE Multilingual
Not directly comparable
SWE Multimodal
Not directly comparable
DeepSWE
Claude Opus 5 leads this result
FrontierCode 1.1 Main
Not directly comparable
FrontierCode 1.1 Extended
Not directly comparable
FrontierSWE v2
Shared sourceClaude Opus 5 leads this result
ProgramBench (episode 1)
Not directly comparable
ProgramBench
Claude Opus 5 leads this result
cursorBench32
Shared sourceClaude Opus 5 leads this result
VulcanBench v3
Claude Opus 5 leads this result
VulcanBench CII v1
Not directly comparable
LiveCodeBench (Vals)
Claude Opus 5 leads this result
SWE-bench (Vals)
Claude Opus 5 leads this result
cursorBench40
Not directly comparable
FrontierSWE
Not directly comparable
Kimi Code Bench v2
Not directly comparable
sweMarathon
Not directly comparable
PostTrain Bench
Not directly comparable
MLS-Bench Lite
Not directly comparable
OpenHarmony Bench
Not directly comparable
HLE
Claude Opus 5 leads this result
HLE w/o tools
Claude Opus 5 leads this result
HLE-Verified
Not directly comparable
LABBench2
Not directly comparable
HealthBench (raw)
Not directly comparable
HealthBench (length-adjusted)
Not directly comparable
HealthBench Professional
Not directly comparable
HealthBench Professional (raw)
Not directly comparable
BioMysteryBench (human-solvable)
Not directly comparable
BioMysteryBench (human-difficult)
Not directly comparable
SpatialBench Verified
Not directly comparable
SingleCellBench
Not directly comparable
ProteinGym Hard
Not directly comparable
Protein Design
Not directly comparable
Organic chemistry V2
Not directly comparable
Protocols (troubleshooting)
Not directly comparable
Protocols (understanding)
Not directly comparable
GPQA Diamond (Vals)
Claude Opus 5 leads this result
MMLU-Pro (Vals)
Claude Opus 5 leads this result
GPQA
Not directly comparable
GPQA-D
Not directly comparable
IMO 2026
Not directly comparable
RiemannBench (no tools)
Not directly comparable
RiemannBench (tools)
Not directly comparable
ArXivMath Jun. 2026 (no tools)
Not directly comparable
ArXivMath Jun. 2026 (tools)
Not directly comparable
Chartography (no tools)
Not directly comparable
Chartography (tools)
Not directly comparable
BenchCAD Vision2Code (no tools)
Not directly comparable
BenchCAD Vision2Code (tools)
Not directly comparable
GDP.pdf (no tools)
Not directly comparable
GDP.pdf (tools)
Not directly comparable
OfficeQA
Not directly comparable
OfficeQA Pro
Claude Opus 5 leads this result
MMMU-Pro
Not directly comparable
MMMU-Pro w/ Python
Not directly comparable
CharXiv w/o tools
Not directly comparable
CharXiv
Not directly comparable
MathVision
Not directly comparable
MathVision w/ Python
Not directly comparable
BabyVision w/ Python
Not directly comparable
ZeroBench
Not directly comparable
ZeroBench w/ Python
Not directly comparable
WorldVQA ForceAnswer
Not directly comparable
OmniDocBench
Not directly comparable
PerceptionBench
Not directly comparable
Claude Opus 5 has the higher public score estimate, 81.89 versus 74.9, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.
Claude Opus 5 leads the public coding lane, 75.9 to 68, with Supported evidence for both models, although the 90% intervals overlap.
Claude Opus 5 leads the public agentic tasks lane, 78.1 to 72, with Supported evidence for both models, although the 90% intervals overlap.
For the stated presets, chat costs $0.0175 on Claude Opus 5 and $0.0105 on Kimi K3; repository review costs $0.325 and $0.195; the cache-heavy agent loop costs $0.45 and $0.27. Costs use the listed standard API rates.
Kimi K3 has the larger documented context window: 1.05M, compared with 1M.
Last updated September 14, 2026
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.