AI models compared

Which are loved

All families

Family	Recommend	Sentiment	Latest release	Praised for	Top caveat
Claude	2,266	34% · 20% of 5,251	Claude Opus 5.581% loved, 202 people · ▲ +39 vs Claude Opus 5	Coding quality, Building tools	Safety flags
GPT	633	35% · 17% of 1,437	GPT Sol 6.185% loved, 45 people · ▲ +25 vs GPT Sol 6	Cost efficiency, Coding	Unreliable output
Gemini	627	33% · 25% of 1,601	Gemini 4 Argon76% loved, 38 people	Flash models, Everyday chat	Accuracy
Qwen	536	47% · 9% of 951	Qwen 3.887% loved, 223 people · ▲ +1 vs Qwen 3.7	Local speed, Coding quality	Overthinking loops
DeepSeek	487	49% · 10% of 878	DeepSeek V4 Flash 4.193% loved, 111 people · ▼ -1 vs DeepSeek V4 Flash	Low cost, Good enough quality	Insecure code
ChatGPT	409	24% · 29% of 1,511	no release with 10+ people	Coding help, Everyday assistant	Answer quality
GGLM	347	51% · 9% of 580	GLM 5.382% loved, 92 people · ▼ -7 vs GLM 5.2	Low cost, Good enough	Verbal tics
KKimi	221	41% · 12% of 471	Kimi 3.169% loved, 22 people · ▼ -9 vs Kimi 3	Cost efficiency, Near-frontier ability	Usage costs
Gemma	165	46% · 10% of 325	Gemma 481% loved, 244 people · ▼ -19 vs Gemma 3	Modest hardware, Writing	Coding quality
Grok	132	26% · 36% of 439	no release with 10+ people	Coding help, Research	Data handling
Mistral	91	35% · 21% of 225	Mistral Large 443% loved, 37 people	Low cost, Local running	Competitiveness
Jev	62	41% · 16% of 147	no release with 10+ people	Cost and speed, Accuracy	Specialized models win
NNVIDIA Parakeet	62	57% · 7% of 101	NVIDIA Parakeet 389% loved, 27 people	Speed, Daily dictation	Multilingual accuracy
MiniMax	55	47% · 13% of 107	MiniMax 391% loved, 19 people · ▲ +6 vs MiniMax 2.7	Cheap token plans, Agent utility roles	Slow locally
NNano Banana	55	43% · 14% of 115	Nano Banana 275% loved, 11 people	Image editing, Space visualization	Output errors
gpt-oss	54	47% · 6% of 106	no release with 10+ people	Runs locally, Speed	20B mistakes
MiMo	51	47% · 11% of 90	MiMo Pro 2.670% loved, 11 people · ▼ -25 vs MiMo Pro 2.5	Low cost, Coding and specs	Overthinking
Muse	47	43% · 16% of 97	Muse Spark 1.383% loved, 11 people	Errand automation, Ongoing monitoring	Privacy
WWhisper	43	35% · 7% of 123	no release with 10+ people	Runs locally, Transcription quality	Outdated
Cursor Composer	41	62% · 18% of 65	Cursor Composer 2.588% loved, 35 people · ▼ -12 vs Cursor Composer 2	Speed, Low cost	Complex tasks

Recommend counts anyone who recommended the family at least once. The bar uses each person's latest stance, so its % can be lower. "Loved" is recommend ÷ (recommend + warn) among people commenting while that release was the newest in its line, starting from its third mention. Model pages show all-time numbers for a version, so they can differ. Praise and caveats cover each of the top 11 families across all its tiers, and are written by a model from cited comments.

Best for

% of people recommending the model for that job, over the number of people who mentioned it for that job. Darker green means more recommend it. Open a cell to read comments.

Family	Coding	Agents	Running locally	Writing	Research and search	Images and video	Speech and audio	Everyday chat
Claude	44%2208	40%349	62%21	39%208	51%276	44%70	55%31	46%188
GPT	55%409	46%65	·	42%40	50%50	29%45	54%28	45%38
Gemini	45%224	36%44	25%8	60%40	53%116	50%100	46%46	44%135
Qwen	68%255	67%63	60%450	89%9	68%22	61%66	49%41	57%21
DeepSeek	75%264	62%74	66%56	44%9	63%19	60%10	·	74%38
ChatGPT	42%127	36%11	·	40%50	47%135	45%38	56%25	31%151
GGLM	71%202	59%37	47%38	·	50%6	50%10	·	83%12
KKimi	56%107	53%19	33%12	62%8	71%7	·	·	82%17
Gemma	59%44	61%18	59%148	73%11	69%13	50%12	50%14	58%12
Grok	54%52	29%7	·	56%9	73%26	27%15	50%10	48%31
Mistral	48%25	·	54%24	57%7	88%8	56%9	48%31	45%20
Jev	·	57%7	·	·	·	·	·	·
NNVIDIA Parakeet	·	·	61%18	·	·	·	58%100	·
MiniMax	65%26	91%11	56%9	·	·	67%9	·	·

The job comes from a keyword match on the use a comment states, plus the category label when it names a job (for example "AI image model"). About 4 in 10 mentions state a usable job, so these are samples of those comments, not of all users. One comment can count under several jobs. Cells with fewer than 5 people are hidden.

Head to head

Comments naming both families, or listing one as the alternative to the other.

1. Claude vs GPT 953 comments
   91 favor Claude151 favor GPT
2. Claude vs Gemini 674 comments
   80 favor Claude64 favor Gemini
3. Claude vs ChatGPT 515 comments
   74 favor Claude34 favor ChatGPT
4. Claude vs DeepSeek 339 comments
   12 favor Claude87 favor DeepSeek
5. Claude vs GLM 332 comments
   25 favor Claude89 favor GLM
6. Gemini vs ChatGPT 274 comments
   53 favor Gemini27 favor ChatGPT
7. GPT vs Gemini 256 comments
   27 favor GPT34 favor Gemini
8. Claude vs Qwen 248 comments
   22 favor Claude61 favor Qwen
9. Claude vs Kimi 246 comments
   21 favor Claude56 favor Kimi
10. DeepSeek vs GLM 195 comments
    17 favor DeepSeek16 favor GLM
11. Qwen vs Gemma 169 comments
    16 favor Qwen18 favor Gemma
12. GPT vs DeepSeek 168 comments
    17 favor GPT33 favor DeepSeek

A comment favors a model when it recommends that one and warns about or only compares the other, or says the person switched to it. Most comments show no clear preference, so the bars compare only the ones that do.

Who's switching

1. Claude → GPT4714 went back the other way
2. Claude → DeepSeek260 went back the other way
3. Claude → GLM251 went back the other way
4. ChatGPT → Claude2311 went back the other way
5. ChatGPT → Gemini207 went back the other way
6. GPT → Claude1447 went back the other way
7. Claude → ChatGPT1123 went back the other way
8. Claude → Kimi110 went back the other way
9. Claude → Gemini106 went back the other way
10. Gemini → ChatGPT720 went back the other way
11. Claude → Qwen70 went back the other way
12. Gemini → Claude610 went back the other way
13. GLM → DeepSeek54 went back the other way
14. Claude → Grok52 went back the other way
15. Qwen → Gemma52 went back the other way
16. DeepSeek → GLM45 went back the other way
17. Kimi → GLM42 went back the other way
18. GPT → DeepSeek42 went back the other way
19. ChatGPT → DeepSeek40 went back the other way
20. Kimi → DeepSeek41 went back the other way

People who said they moved from one family to another, 337 in all. It's a small sample of people who chose to say so.