Compare AI models
| Specifications | ||
ParametersThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality. | 1T | 49B |
Context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider. | 256k | 128k |
| API pricingUSD per 1M tokens · lower wins | ||
Cheapest inputLowest input rate across every provider on OpenRouter, the lab included. The cheapest endpoint may run a quantised build or a shorter context — see "Available from" on the model page. | $0.49AtlasCloud | — |
Cheapest outputLowest output rate across every provider on OpenRouter, the lab included. May come from a different provider than the cheapest input. | $2.50AtlasCloud | — |
| $223AtlasCloud · int4 | — | |
| Benchmarks | ||
LiveCodeBenchCompetitive coding — Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better. | 85% | 73.58% |
AIME 2025Reasoning — single-answer accuracy on the 2025 invitational mathematics exam without tools | 96.1% | 82.71% |
GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. | 87.6% | 71.97% |
MMLU-ProKnowledge — accuracy on the ten-option MMLU-Pro benchmark | 87.1% | 79.53% |
BullshitBench v2Nonsense detection — Given a confidently-worded but nonsensical prompt, does the AI spot that it makes no sense and push back — instead of playing along and inventing an answer? The score is how often it clearly called out the nonsense. Higher is better. | 52% | — |
SWE-Bench ProAgentic coding — Can the AI fix real bugs in real software? It's handed actual problems from open-source projects and has to write code that genuinely solves them. Higher is better. | 50.7% | — |
SWE-Bench VerifiedCoding — Real coding tasks pulled from open-source projects — the AI has to find and fix actual bugs. A human-checked version of the original SWE-Bench. Higher is better. | 76.8% | — |
SWE-Bench MultilingualMultilingual coding — Like SWE-Bench, but the coding problems span many programming languages, not just one. Tests how broadly the AI can code. Higher is better. | 73% | — |
Next.js EvalsNext.js coding — Vercel's open eval of how well AI coding agents build and migrate real Next.js apps — measured as the share of tasks the agent completes successfully. Higher is better. | 18% | — |
Terminal-Bench 2.0Agentic terminal coding — Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? (Version 2.0 of the test.) Higher is better. | 50.8% | — |
BrowseCompWeb browsing — Can the AI browse the web and track down hard-to-find answers? Higher is better. | 60.6% | — |
Humanity's Last Exam · no toolsMultidisciplinary reasoning — Humanity's Last Exam — extremely hard expert questions across many subjects, written so you can't just look up the answer. “No tools” means the AI answers on its own. Higher is better. | 30.1% | — |
Humanity's Last Exam · text only, no toolsMultidisciplinary reasoning — accuracy on the 2,158 text-only questions without tools | — | 7.64% |
Humanity's Last Exam · with toolsMultidisciplinary reasoning — Humanity's Last Exam — extremely hard expert questions across many subjects. “With tools” means the AI is allowed to search the web or run code while answering. Higher is better. | 50.2% | — |
MATH-500Reasoning — accuracy on the 500-problem MATH subset | — | 97.4% |
AIME 2024Reasoning — single-answer accuracy on the 2024 invitational mathematics exam without tools | — | 87.5% |
OSWorld-VerifiedAgentic computer use — Can the AI actually operate a computer — clicking, typing, and using real apps — to finish tasks on its own? Higher is better. | 63.3% | — |
| Overview | ||
| Company | Moonshot AI | NVIDIA |
| Release date | Jan 27 2026 | Jul 25 2025 |
| Access | Open Weight | Open Weight |
| Model details | View model | View model |
Other comparisons
Kimi K2.5vsClaude Fable 5.2Llama Nemotron Super 49B v1.5vsClaude Fable 5.2Kimi K2.5vsGPT-6.2 AstraLlama Nemotron Super 49B v1.5vsGPT-6.2 AstraKimi K2.5vsGemini 4 NeonLlama Nemotron Super 49B v1.5vsGemini 4 NeonKimi K2.5vsMuse EmberLlama Nemotron Super 49B v1.5vsMuse EmberKimi K2.5vsGrok 4.7Llama Nemotron Super 49B v1.5vsGrok 4.7Kimi K2.5vsDeepSeek-V4.1-Flash-1017Llama Nemotron Super 49B v1.5vsDeepSeek-V4.1-Flash-1017Frequently asked questions
Kimi K2.5 leads Llama Nemotron Super 49B v1.5 on 4 of the 4 benchmarks they both report (LiveCodeBench, AIME 2025, GPQA Diamond, MMLU-Pro). Llama Nemotron Super 49B v1.5 shipped 186 days before Kimi K2.5, so benchmark comparisons should account for the intervening progress.
Kimi K2.5 has 1T parameters, while Llama Nemotron Super 49B v1.5 has 49B. Context windows are 256k (Kimi K2.5) vs 128k (Llama Nemotron Super 49B v1.5).
On LiveCodeBench, Kimi K2.5 leads at 85% vs Llama Nemotron Super 49B v1.5 at 73.58%. On AIME 2025, Kimi K2.5 leads at 96.1% vs Llama Nemotron Super 49B v1.5 at 82.71%. On GPQA Diamond, Kimi K2.5 leads at 87.6% vs Llama Nemotron Super 49B v1.5 at 71.97%. On MMLU-Pro, Kimi K2.5 leads at 87.1% vs Llama Nemotron Super 49B v1.5 at 79.53%.