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. | — | 16B |
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. | 32k | 8k |
| Benchmarks | ||
HumanEvalFunction synthesis — 164 small Python problems: the AI is given a function's description and has to write the working function. This was the coding benchmark of the GPT-3.5 and GPT-4 era, before the field moved to fixing real bugs in real repositories. Higher is better. | 74.4% | 48.1% |
MATHReasoning — accuracy on the original competition mathematics test set | 53.2% | 45.3% |
MMLUGeneral knowledge — A 57-subject multiple-choice exam — history, law, medicine, maths — that was the standard measure of how much a model knows from 2020 until roughly 2024, when frontier scores crowded into the high 80s and labs moved on to harder tests. The scores here were published years apart under different testing setups, so read them as a historical record rather than a like-for-like ranking. Higher is better. | 83.7% | 70% |
GSM8KGrade-school math — Grade-school maths word problems that take a few steps of arithmetic to work through. It separated the models of 2022 and 2023 sharply, then saturated. One caveat on the historical numbers: OpenAI included part of the GSM8K training set in GPT-4's pre-training mix, so GPT-4's score is not a clean few-shot result. Higher is better. | 88.9% | 77.4% |
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. | — | 14.3% |
HellaSwagReasoning — commonsense completion of short scenarios | 87.8% | — |
WinoGrandeReasoning — resolving ambiguous pronouns in context | — | 74.6% |
GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. | — | 35.2% |
SimpleQAKnowledge — accuracy on short, fact-seeking questions in the original SimpleQA set | — | 5.6% |
MMLU-ProKnowledge — accuracy on the ten-option MMLU-Pro benchmark | — | 42.4% |
MGSMMultilingual math — solving grade-school word problems translated into multiple languages | 79% | — |
MathVistaMultimodal reasoning — mathematical reasoning about visual information | 53% | — |
MMMUMultimodal — Tests the AI on understanding images and text together across many college subjects. Higher is better. | 59.4% | — |
| Overview | ||
| Company | Moonshot AI | |
| Release date | Feb 8 2024 | Feb 22 2025 |
| Access | Closed | Open Weight |
| Model details | View model | View model |
Other comparisons
Gemini 1.0 UltravsClaude Fable 5.2MoonlightvsClaude Fable 5.2Gemini 1.0 UltravsGPT-6.2 AstraMoonlightvsGPT-6.2 AstraGemini 1.0 UltravsMuse EmberMoonlightvsMuse EmberGemini 1.0 UltravsGrok 4.7MoonlightvsGrok 4.7Gemini 1.0 UltravsDeepSeek-V4.1-Flash-1017MoonlightvsDeepSeek-V4.1-Flash-1017Gemini 1.0 UltravsMistral Large 4MoonlightvsMistral Large 4Frequently asked questions
Gemini 1.0 Ultra leads Moonlight on 4 of the 4 benchmarks they both report (HumanEval, MATH, MMLU, GSM8K). Gemini 1.0 Ultra shipped 380 days before Moonlight, so benchmark comparisons should account for the intervening progress.
Context windows are 32k (Gemini 1.0 Ultra) vs 8k (Moonlight). Gemini 1.0 Ultra is closed, while Moonlight is open weight.
On HumanEval, Gemini 1.0 Ultra leads at 74.4% vs Moonlight at 48.1%. On MATH, Gemini 1.0 Ultra leads at 53.2% vs Moonlight at 45.3%. On MMLU, Gemini 1.0 Ultra leads at 83.7% vs Moonlight at 70%. On GSM8K, Gemini 1.0 Ultra leads at 88.9% vs Moonlight at 77.4%.