🆕 NewJuly 16, 2026
Kimi K3 released
Moonshot AI launches its 2.8T-parameter flagship with 1M token context, native vision, and always-on reasoning. Open weights promised for late July.
Kimi K3
🧠 Reasoning🌐 Web Search
Moonshot AI's brand-new 2.8-trillion-parameter flagship. Frontier scores on agentic and reasoning work, with open weights promised within weeks.
Released July 16, 2026
Pricing
Input tokens$3.00/M
Output tokens$15.00/M
Capacity
Context window1.0M tokens
Capabilities
✓Reasoning & Planning
✓Web Search
✗Open Source
Best Scores
GDPval-AA v21686.0 pts
GPQA Diamond93.5%
SWE-bench Verified93.4%
Terminal-Bench 2.188.3%
Artificial Analysis Intelligence Index57.0 pts
Best For
💻coding
🔬research
📊analysis
Benchmark Scores
Specialized Skills
GDPval-AA v2
Real paid work from 44 different jobs, such as law, nursing, and software. Judges compare two answers side by side without knowing which model wrote them, and the winner gains rating points. A typical human expert scores 1000, so a higher number means the work was picked over a human more often.
Independently measured by Artificial Analysis.
Knowledge
GPQA Diamond
PhD-level science questions written so you cannot just Google the answer. Tests whether the model can reason about hard science.
Provider-reported; no independent run recorded yet.
Humanity's Last Exam
PhD-level questions across many subjects. Tests deep reasoning on the hardest questions humans can ask.
Provider-reported; no independent run recorded yet.
Software Engineering
SWE-bench Verified
Hands the AI bugs from actual software projects and counts how many it fixes. Like a coding job interview, but with real work.
Independently measured by Vals AI (mini-swe-agent).
Terminal-Bench 2.1
Puts the AI in front of a computer terminal and asks it to finish multi-step tasks on its own. Measures how good an "AI agent" it is.
Provider-reported; an independent run by Vals AI (Terminus 2) lands at 80.9%.
Reasoning
Artificial Analysis Intelligence Index
A frequently refreshed overall score made from nine modern tests: real work, tool use, terminal tasks, science, hard questions, and long-context reasoning. It is a scorecard rather than a percent correct.
Independently measured by Artificial Analysis.
Why Choose Kimi K3?
Kimi K3 posts some of the strongest published agentic numbers of any model, closed or open, and Moonshot has promised to release the weights, a rare combination of frontier capability now and self-hosting later. The 1M context and native vision make it a strong pick for long-horizon tool-use agents.
How It Compares
vs GPT-5.6 Sol
K3 is cheaper and matches Sol's published terminal scores; Sol has independent verification and a mature ecosystem.
vs GLM-5.2
K3 posts higher published agentic scores; GLM-5.2 is already downloadable and MIT-licensed today.