Kimi K3
Moonshot AI · released
Moonshot's flagship: a 2.8-trillion-parameter open-weight model with native image and video understanding and a 1M-token context, built for long autonomous coding sessions, research-style work and deep reasoning. It costs far more than earlier Kimi models.
Best for: Largest open-weight model · Long-context reasoning · Research-style work
- Intelligence Index
- 43.6
- #13 of 38 · at Max effort
- Price per 1M tokens
- $3 in $15 out
- Blended $6 at 3:1
- Context window
- 1M
- Up to 1M output
- Output speed
- 42 tokens/s
- $2 measured cost per task
Capability profile
One 0–100 score per area, placing this model between the weakest and strongest tracked model across that area's benchmarks.
- Overall69 · #11 of 38
Based on 3 of 3 benchmarks
- Coding48 · #24 of 39
Based on 5 of 5 benchmarks
- Agents & knowledge work70 · #12 of 40
Based on 6 of 6 benchmarks
- Reasoning & science73 · #10 of 39
Based on 2 of 2 benchmarks
- Knowledge & reliability83 · #9 of 39
Based on 2 of 2 benchmarks
- Vision77 · #10 of 31
Based on 1 of 2 benchmarks
Reasoning effort trade-off
The same model at each effort setting. Lower settings give up a few points for a fraction of the cost and wait.
| Setting | Intelligence Index | Cost / task | First answer |
|---|---|---|---|
| Max | 43.6 | $2 | 51s |
| Low | 30.1 | $1.15 | 53s |
Measured by Artificial Analysis. First answer is the median wait until the first answer token, including thinking time.
Benchmark results
All third-party results on record. Rank is among the tracked models that have been tested on that benchmark.
Overall
#13 of 38· Max effort· Artificial Analysis
- Vals Index50.3%
#17 of 26· Max effort· $6.38 per test· Vals AI
- Arena · Text1488 ±5
#6 of 33· Max effort· 28,280 votes· Arena
Coding
- Terminal-Bench 4.012.6%
#22 of 38· Max effort· Artificial Analysis
- SciCode59.5%
#6 of 38· Max effort· Artificial Analysis
- Arena · WebDev1655 ±6
#9 of 34· Max effort· 17,061 votes· Arena
- Vibe Code Bench85.0%
#11 of 26· Max effort· Vals AI
- Code Migration16.0%
#24 of 26· Max effort· Vals AI
Agents & knowledge work
- GDPval-AA51.8
#18 of 39· Max effort· Artificial Analysis
- Arena · Agent+4.2 ±0.5
#7 of 29· Max effort· Arena
- Finance Agent v253.1%
#16 of 26· Max effort· Vals AI
#11 of 26· Max effort· Vals AI
- Legal Research Bench46.2%
#10 of 26· Max effort· Vals AI
- Tax Agent Bench68.7%
#7 of 26· Max effort· Vals AI
Reasoning & science
- Humanity's Last Exam46.9%
#11 of 39· Max effort· Artificial Analysis
- CritPt23.4%
#10 of 39· Max effort· Artificial Analysis
- ARC-AGI-2saturating60.4%
#12 of 16· Max effort· $1.59 per task· ARC Prize
- ARC-AGI-3Not tested
- GPQA Diamondsaturating93.5%
#5 of 27· Max effort· Artificial Analysis
Knowledge & reliability
#11 of 39· Max effort· Artificial Analysis
- AA-LCR88.7%
#1 of 39· Max effort· Artificial Analysis
Vision
- MMMU-Pro80.5%
#9 of 27· Max effort· Artificial Analysis
- Arena · VisionNot tested
Reported by Moonshot AI
Numbers the lab published at launch, run on its own harness and settings. Useful context, but not independently verified and never used in the rankings on this site.
- DeepSWE v1.167.5%
- MCP-Atlas · public subset84.2%
- BrowseComp · with context compaction91.2%
- GPQA Diamond93.5%
Specifications
- Lab
- Moonshot AI
- Released
- Jul 16, 2026
- Availability
- Generally available
- Context window
- 1,048,576 tokens
- Max output
- 1,048,576 tokens
- Accepts
- Text, Images, Video
- Knowledge cutoff
- Not published
- Weights
- Open
- Licence
- Kimi K3 License
- Parameters
- 2.8T total · 104B active
- Reasoning control
- Always on · effort low to max
- API model ID
kimi-k3
Pricing
- Input / 1M tokens
- $3
- Cached input / 1M
- $0.30
- Output / 1M tokens
- $15
- Blended (3:1)
- $6
- Measured cost / task
- $2
Worth knowing
- The weights use a custom licence rather than a standard open-source one.
Official sources
Closest alternatives
The models nearest to this one on overall score.
23 benchmarks tracked. See methodology and sources.