AA-OmniscienceVerified
47.6%
AA-Omniscience Accuracy
Models / Moonshot
Kimi K3
Kimi K3 · Released 2026-07-16
Moonshot AI's 2.8T-parameter multimodal reasoning model for long-horizon coding and knowledge work, with a 1M-token context window, public API access, and open weights on Hugging Face.
Open-frontier option for large-repository coding, tool use, visual reasoning, and context-heavy research workflows — via API or self-hosted weights.
Composite58
Context1M
Input / 1M$3.0
Output / 1M$15
Knowledge cutoffunpublished
Output speed42 t/s
Statussolid
Family profile
Best published score in each covered benchmark family.
76/100 avg
- Knowledge
- 48
- Reasoning
- 94
- Coding
- 60
- Agentic
- 85
- Multimodal
- 81
- Long context
- 89
6 tested benchmark families
Benchmark placements
Per-benchmark positions
Where Kimi K3 places on each public benchmark source that publishes a row. Rank counts every model with a latest row in the same test — not a universal quality score.
| Benchmark | Family | Rank | Score | Source | Date |
|---|---|---|---|---|---|
| AA-LCR | Long context | #1/ 184 | 89 | Artificial Analysis | 2026-09-28 |
| SciCode | Coding | #5/ 97 | 60 | Artificial Analysis | 2026-09-28 |
| Terminal-Bench-Science | Agentic | #7/ 9 | 7 | Terminal-Bench-Science 0.1 announcement | 2026-08-27 |
| MMMU-Pro | Multimodal | #8/ 8 | 81 | Artificial Analysis — MMMU-Pro | 2026-09-01 |
| CritPt | Reasoning | #9/ 12 | 23 | Artificial Analysis — CritPt | 2026-09-01 |
| AA-Omniscience | Knowledge | #11/ 12 | 48 | Artificial Analysis — AA-Omniscience | 2026-09-01 |
| GPQA Diamond | Reasoning | #11/ 184 | 94 | Artificial Analysis | 2026-09-28 |
| Terminal-Bench | Agentic | #14/ 167 | 85 | Artificial Analysis | 2026-09-28 |
| Humanity's Last Exam | Reasoning | #17/ 186 | 47 | Artificial Analysis | 2026-09-28 |
| AA output speed | Performance | #134/ 195 | Artificial Analysis | 2026-10-12 |
Family context
Strongest rows in each family
The three highest-scoring models with pages in each capability family. Where Kimi K3 shows up, it's highlighted.
KnowledgeKimi K3 · 48
- 167
- 265
- 361
ReasoningKimi K3 · 94
- 196
- 295
- 395
Math
- 152
- 252
- 350
CodingKimi K3 · 60
- 174
- 273
- 370
AgenticKimi K3 · 85
- 191
- 289
- 389
MultimodalKimi K3 · 81
- 186
- 285
- 384
Long contextKimi K3 · 89
- 189
- 286
- 385
Newest receipts
10 recentLast 120 days
Benchmark rows added to the public ledger for Kimi K3 in the last 120 days. Older rows live in the full table below.
| Benchmark | Family | Score | Source | Days ago |
|---|---|---|---|---|
| AA output speed | Performance | 42 t/s | Artificial Analysis | 10 |
| AA-LCR | Long context | 88.67% | Artificial Analysis | 24 |
| GPQA Diamond | Reasoning | 93.5% | Artificial Analysis | 24 |
| Humanity's Last Exam | Reasoning | 46.9% | Artificial Analysis | 24 |
| SciCode | Coding | 59.5% | Artificial Analysis | 24 |
| Terminal-Bench | Agentic | 85.02% | Artificial Analysis | 24 |
| CritPt | Reasoning | 23.4% | Artificial Analysis — CritPt | 51 |
| AA-Omniscience | Knowledge | 47.6% | Artificial Analysis — AA-Omniscience | 51 |
| MMMU-Pro | Multimodal | 80.5% | Artificial Analysis — MMMU-Pro | 51 |
| Terminal-Bench-Science | Agentic | 7.1% | Terminal-Bench-Science 0.1 announcement | 56 |
Family coverage
Best score per family
A high composite that hides a weak family is a trap. These bars surface the families where this model hasn't been publicly tested, and where it leads.
Price vs. performance
Where Kimi K3 sits on the scatter
Hover any dot for name, score, and input price. Keyboard: tab through the top twelve, or use the ranking below.
AnthropicOpenAIGoogleAlibabaMetaZhipuxAIDeepSeekMiniMaxXiaomiMoonshotOtherMeituanMistralNVIDIACohere
Full benchmark ledger
Every catalog benchmark for Kimi K3. Scores link to the original source; gaps mean no public row exists yet.
10 of 13 catalog benchmarks have a sourced row for Kimi K3.
Knowledge
Reasoning
CritPtVerified
23.4%
CritPt composite (AA run)
GPQA DiamondVerified
93.5%
GPQA Diamond (AA run)
Humanity's Last ExamVerified
46.9%
HLE (AA run)
Coding
SciCodeVerified
59.5%
SciCode (AA run)
Agentic
Terminal-BenchVerified
85.02%
Terminal-Bench v2.1 (AA run)
Terminal-Bench-ScienceVerified
7.1%
TB-Science 0.1 resolution (Claude Code)
Multimodal
MMMU-ProVerified
80.5%
MMMU-Pro (AA run)
Long context
AA-LCRVerified
88.67%
AA-LCR (AA run)
Performance
AA output speedVerified
42 t/s
Median output tokens/s (1k prompt, default provider)
Changelog
Version history
- Amazon Bedrock, US and global profiles. Kimi API list unchanged.
Sources
Where the numbers come from
- Kimi K3 launch and technical overview· checked 2026-07-27
- Kimi K3 open weights on Hugging Face· checked 2026-07-27
- Kimi K3 API pricing· checked 2026-07-27
- Artificial Analysis model record· checked 2026-07-27
- Amazon Bedrock Kimi K3· checked 2026-09-23
Questions this model page answers
Every answer below is assembled from the dated fields on this page. Nothing is written separately for search.
What is Kimi K3?
Kimi K3 is a Moonshot model: Moonshot AI's 2.8T-parameter multimodal reasoning model for long-horizon coding and knowledge work, with a 1M-token context window, public API access, and open weights on Hugging Face.
What is Kimi K3 for?
Open-frontier option for large-repository coding, tool use, visual reasoning, and context-heavy research workflows — via API or self-hosted weights.
What does Kimi K3 cost?
Kimi K3 lists $3.0 input and $15 output per 1M tokens. Pricing last checked 2026-09-23.
How large is the Kimi K3 context window?
Kimi K3 publishes a 1M token context window.
What is the VerdictPal status of Kimi K3?
Kimi K3 carries Solid status on the model atlas.
The Pack · Editorial newsletter
Know when the rules move.
One short email when a university changes its AI rule for theses, or a research tool changes its pricing or privacy terms. No tracking, no third-party analytics. Unsubscribe in one click.