Agentic
Directional only- Kimi K3
- 71.9
- Supported · #4/152
- Mercury 2.5
- 44.5
- Estimated · #91/152
- Basis
- BenchAlign lane · 11 vs 2 public rows
- Reading
- Directional only
Kimi K3 vs Mercury 2.5
Updated September 8, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.
Decision reading
At least one model is not scored in the current public ranking lane, so the page does not name an overall quality winner.
2 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
Kimi K3
Kimi K3 has the larger documented context window.
Confidence: documented
Chat turn cost
1K fresh input + 500 output tokens
Mercury 2.5
Mercury 2.5 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
Mercury 2.5
Mercury 2.5 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Show secondary and unsupported calls
Repository review cost
50K fresh input + 3K output tokens
Mercury 2.5
Mercury 2.5 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
Mercury 2.5 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
Mercury 2.5 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.
41 Kimi K3 only
2 Shared
3 Mercury 2.5 only
- Shared results
- 2
- Kimi K3 only
- 41
- Mercury 2.5 only
- 3
- Like-for-like categories
- 0 / 8
3 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- Kimi K3
- 68.0
- Supported · #6/151
- Mercury 2.5
- 47.8
- Estimated · #73/151
- Basis
- BenchAlign lane · 13 vs 1 public rows
- Reading
- Directional only
Knowledge
Directional only- Kimi K3
- 72.1
- Supported · #8/182
- Mercury 2.5
- 48.3
- Estimated · #94/182
- Basis
- BenchAlign lane · 6 vs 1 public rows
- Reading
- Directional only
Reasoning
Not comparable- Kimi K3
- 78.5
- #3/18
- Mercury 2.5
- 67.7
- Unranked · 1 rankable row
- Basis
- Provisional lane · 0 vs 0 weighted rows
- Reading
- Not comparable
Math
Not comparable- Kimi K3
- Not ranked
- Mercury 2.5
- Not ranked
- Basis
- Provisional lane · 0 vs 0 weighted rows
- Reading
- Not comparable
Multilingual
Not comparable- Kimi K3
- Not ranked
- Mercury 2.5
- Not ranked
- Basis
- Provisional lane · 0 vs 0 weighted rows
- Reading
- Not comparable
Multimodal
Not comparable- Kimi K3
- 89.5
- #1/48
- Mercury 2.5
- Not ranked
- Basis
- Provisional lane · 3 vs 0 weighted rows
- Reading
- Not comparable
Instruction following
Not comparable- Kimi K3
- Not ranked
- Mercury 2.5
- 80.1
- #53/121
- Basis
- Provisional lane · 0 vs 1 weighted rows
- Reading
- Not comparable
| Category | Kimi K3 | Mercury 2.5 | Basis | Reading |
|---|---|---|---|---|
| Agentic | 71.9Supported · #4/152 | 44.5Estimated · #91/152 | Directional onlyBenchAlign lane · 11 vs 2 public rows | Directional only |
| Coding | 68.0Supported · #6/151 | 47.8Estimated · #73/151 | Directional onlyBenchAlign lane · 13 vs 1 public rows | Directional only |
| Knowledge | 72.1Supported · #8/182 | 48.3Estimated · #94/182 | Directional onlyBenchAlign lane · 6 vs 1 public rows | Directional only |
| Reasoning | 78.5#3/18 | 67.7Unranked · 1 rankable row | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| 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 |
| Multimodal | 89.5#1/48 | Not ranked | Not comparableProvisional lane · 3 vs 0 weighted rows | Not comparable |
| Instruction following | Not ranked | 80.1#53/121 | Not comparableProvisional lane · 0 vs 1 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.
A shared-evidence shape is not available.
BenchLM does not draw a radar or infer missing axes when the matched evidence is too sparse.
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
- Kimi K3
- $0.0105
- Fits in one request
- Mercury 2.5
- $0.00012
- Fits in one request
Mercury 2.5 has the lower modeled cost
Costs use the listed standard API rates.
Repository review
50K fresh input + 3K output tokens
- Kimi K3
- $0.195
- Fits in one request
- Mercury 2.5
- $0.00245
- Fits in one request
Mercury 2.5 has the lower modeled cost
Costs use the listed standard API rates.
Cache-heavy agent loop
200K cached + 20K fresh input + 10K output tokens
- Kimi K3
- $0.27
- Fits in one request
- Mercury 2.5
- $0.0031
- Fits in one request
Mercury 2.5 has the lower modeled cost
Costs use the listed standard API rates.
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationKimi K3Mercury 2.5
Context window
Maximum documented context; output-token limits may be lower.
Kimi K3
1.05M
Mercury 2.5
API model ID
Kimi K3
Not sourced
Mercury 2.5
Cached-input rate
A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
Kimi K3
$0.3 per 1M cached input tokens
Mercury 2.5
$0.004 per 1M cached input tokens
Inception models and Mercury 2.5 launch pricingDocumented inputs
Kimi K3
Not sourced
Mercury 2.5
Not sourced
Documented outputs
Kimi K3
Not sourced
Mercury 2.5
Not sourced
Provider availability
Kimi K3
Not sourced
Mercury 2.5
Not sourced
Reasoning profile
Kimi K3
Reasoning
Mercury 2.5
Reasoning
Weight access
Kimi K3
Pending
Mercury 2.5
Proprietary
License
Kimi K3
Pending
Mercury 2.5
Proprietary
Release date
Kimi K3
2026-07-16
Mercury 2.5
2026-09-08
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
- At least one model is not scored in the current public ranking lane, so the page does not name an overall quality winner.
- Workload cost
- Repository review: $0.195 vs $0.00245. Cache-heavy agent loop: $0.27 vs $0.0031.
- Context tradeoff
- Kimi K3 has the larger documented window (1.05M).
Run the same representative tasks against both endpoints before changing production traffic.
Benchmark evidence
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Browse raw public benchmark evidence46 rows
Agentic
Terminal-Bench 2.0
Kimi K388.3%SourceMercury 2.5—Not directly comparable
BrowseComp
Kimi K391.2%SourceMercury 2.5—Not directly comparable
DeepSearchQA
Kimi K395.0%SourceMercury 2.534.0%SourceKimi K3 leads this result
Toolathlon-Verified
Kimi K373.2%SourceMercury 2.5—Not directly comparable
MCP Atlas
Kimi K384.2%SourceMercury 2.5—Not directly comparable
AutomationBench
Kimi K330.8%SourceMercury 2.5—Not directly comparable
JobBench
Kimi K352.9%SourceMercury 2.5—Not directly comparable
APEX-Agents
Kimi K337.6%SourceMercury 2.5—Not directly comparable
SpreadsheetBench 2
Kimi K334.8%SourceMercury 2.5—Not directly comparable
DECK-Bench
Kimi K373.5%SourceMercury 2.5—Not directly comparable
Terminal-Bench 2.1 (Vals)
Kimi K380.9%SourceMercury 2.5—Not directly comparable
τ³-bench results
Kimi K3—Mercury 2.596.0%SourceNot directly comparable
Coding
deepSwe
Kimi K367.5%SourceMercury 2.5—Not directly comparable
cursorBench32
Kimi K360.8%SourceMercury 2.5—Not directly comparable
FrontierSWE
Kimi K381.2%SourceMercury 2.5—Not directly comparable
ProgramBench
Kimi K377.8%SourceMercury 2.5—Not directly comparable
Kimi Code Bench v2
Kimi K372.9%SourceMercury 2.5—Not directly comparable
sweMarathon
Kimi K342%SourceMercury 2.5—Not directly comparable
PostTrain Bench
Kimi K336.6%SourceMercury 2.5—Not directly comparable
MLS-Bench Lite
Kimi K348.3%SourceMercury 2.5—Not directly comparable
VulcanBench v3
Kimi K373.7%SourceMercury 2.5—Not directly comparable
OpenHarmony Bench
Kimi K357.3%SourceMercury 2.5—Not directly comparable
FrontierSWE v2
Kimi K325.9%SourceMercury 2.5—Not directly comparable
LiveCodeBench (Vals)
Kimi K387.2%SourceMercury 2.5—Not directly comparable
SWE-bench (Vals)
Kimi K393.4%SourceMercury 2.5—Not directly comparable
SciCode
Kimi K3—Mercury 2.538%SourceNot directly comparable
Knowledge
GPQA
Kimi K393.5%SourceMercury 2.5—Not directly comparable
GPQA-D
Kimi K393.5%SourceMercury 2.579.0%SourceKimi K3 leads this result
HLE
Kimi K356%SourceMercury 2.5—Not directly comparable
HLE w/o tools
Kimi K343.5%SourceMercury 2.5—Not directly comparable
GPQA Diamond (Vals)
Kimi K392.9%SourceMercury 2.5—Not directly comparable
MMLU-Pro (Vals)
Kimi K388.0%SourceMercury 2.5—Not directly comparable
Multimodal
OfficeQA Pro
Kimi K363.3%SourceMercury 2.5—Not directly comparable
MMMU-Pro
Kimi K381.6%SourceMercury 2.5—Not directly comparable
MMMU-Pro w/ Python
Kimi K383.4%SourceMercury 2.5—Not directly comparable
CharXiv w/o tools
Kimi K384.8%SourceMercury 2.5—Not directly comparable
CharXiv
Kimi K391.3%SourceMercury 2.5—Not directly comparable
MathVision
Kimi K394.3%SourceMercury 2.5—Not directly comparable
MathVision w/ Python
Kimi K397.8%SourceMercury 2.5—Not directly comparable
BabyVision w/ Python
Kimi K385.7%SourceMercury 2.5—Not directly comparable
ZeroBench
Kimi K323.0%SourceMercury 2.5—Not directly comparable
ZeroBench w/ Python
Kimi K341.0%SourceMercury 2.5—Not directly comparable
WorldVQA ForceAnswer
Kimi K351.0%SourceMercury 2.5—Not directly comparable
OmniDocBench
Kimi K391.1%SourceMercury 2.5—Not directly comparable
PerceptionBench
Kimi K358.5%SourceMercury 2.5—Not directly comparable
Instruction following
IFBench
Kimi K3—Mercury 2.577%SourceNot directly comparable
Frequently asked questions
Which is better, Kimi K3 or Mercury 2.5?
At least one model is not scored in the current public ranking lane, so the page does not name an overall quality winner. The page therefore keeps the decision tied to the specific documented workload.
Which is better for coding, Kimi K3 or Mercury 2.5?
Kimi K3 scores higher for coding on the public lane, 68 to 47.8. Mercury 2.5 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, Kimi K3 or Mercury 2.5?
Kimi K3 scores higher for agentic tasks on the public lane, 71.9 to 44.5. Mercury 2.5 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, Kimi K3 or Mercury 2.5?
For the stated presets, chat costs $0.0105 on Kimi K3 and $0.00012 on Mercury 2.5; repository review costs $0.195 and $0.00245; the cache-heavy agent loop costs $0.27 and $0.0031. Costs use the listed standard API rates.
Which has the larger context window, Kimi K3 or Mercury 2.5?
Kimi K3 has the larger documented context window: 1.05M, compared with 260K.
Related comparisons
Last updated September 8, 2026
Watch Kimi K3 vs Mercury 2.5
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.