Is AI Dumber Today? Community-tracked AI experience, updated hourly

Exact models

Exact model atlas

Browse community experience for specific model versions.

Updated 2026-08-19T14:46:11Z UTC

Generic family mentions remain in family scores and are not assigned to an exact model.

Exact model directory

89 results
Model Score 14-day trend Mentions Confidence
Claude 12 models
Opus 4.8 68.8+29.9 n=24 Medium
Claude Opus 4.6 61.5+22.6 n=6 Low
Claude Opus 4.5 56.0+17.1 n=6 Low
Claude Opus 5 53.8+14.9 n=33 High
Claude Fable 5 52.7+13.8 n=145 High
Claude Opus 4.8 46.7+7.8 n=6 Low
Sonnet 5 36.5−2.4 n=16 Medium
Claude Haiku 4.5 35.9−3.1 n=6 Low
Claude Sonnet 5 mentions only n=4 not enough data
Claude Sonnet 4.6 mentions only n=2 not enough data
Haiku 4.5 mentions only n=2 not enough data
Claude Opus 4 mentions only n=1 not enough data
DeepSeek 5 models
DeepSeek V4 Flash 72.1+26.3 n=380 High
R1 65.8+20.0 n=7 Low
DeepSeek V4 Pro 33.2−12.6 n=854 High
DeepSeek V3 mentions only n=2 not enough data
DeepSeek V3.2 mentions only n=1 not enough data
Qwen 18 models
Qwen3.6 27B 70.5+9.5 n=76 High
Qwen3.6 35B A3B 69.8+8.8 n=43 High
Qwen3.5-9B 64.1+3.1 n=10 Medium
Qwen3.8 27B 60.8−0.2 n=530 High
Qwen3.8 Max 58.5−2.5 n=29 Medium
Qwen3.8 2.4T A95B 50.6−10.4 n=11 Medium
Qwen3.5-122B-A10B 46.4−14.7 n=8 Low
Qwen3 30B A3B mentions only n=3 not enough data
Qwen3 Coder Next mentions only n=3 not enough data
Qwen-Max mentions only n=2 not enough data
Qwen3 8B mentions only n=2 not enough data
Qwen3 Max mentions only n=2 not enough data
Qwen3.5-4B mentions only n=2 not enough data
Qwen3.7 Plus mentions only n=2 not enough data
Qwen3 32B mentions only n=1 not enough data
Qwen3 235B A22B mentions only n=1 not enough data
Qwen3.5-35B-A3B mentions only n=1 not enough data
Qwen3.8 Max Preview mentions only n=1 not enough data
GPT 18 models
GPT-5.6 Luna 64.9+8.0 n=52 High
GPT-5.4 62.3+5.4 n=7 Low
GPT-5.4 Image 2 61.2+4.3 n=5 Low
gpt-oss-120b 59.6+2.6 n=5 Low
GPT-4o 52.7−4.3 n=9 Low
GPT-5.5 51.5−5.4 n=18 Medium
GPT-5.6 Sol 47.5−9.4 n=182 High
GPT-5.3-Codex-Spark 40.2−16.7 n=6 Low
GPT-4 mentions only n=4 not enough data
GPT-5.6 Sol Pro mentions only n=4 not enough data
gpt-oss-20b mentions only n=3 not enough data
GPT-3.5 Turbo mentions only n=2 not enough data
GPT-5 mentions only n=2 not enough data
GPT-5.6 Terra mentions only n=2 not enough data
Astra mentions only n=1 not enough data
GPT-3 mentions only n=1 not enough data
GPT-5 Mini mentions only n=1 not enough data
GPT-5.3-Codex mentions only n=1 not enough data
Grok 6 models
Grok 4.6 67.0+31.1 n=79 High
Grok 4.5 64.0+28.1 n=24 Medium
Grok Imagine Image 2.0 mentions only n=3 not enough data
Grok 3 mentions only n=2 not enough data
Grok 4 mentions only n=1 not enough data
Grok 4.20 mentions only n=1 not enough data
Gemini 7 models
Gemini 3.7 Flash 78.0+15.7 n=68 High
Gemini 3.6 Flash 51.4−10.9 n=6 Low
Gemini 3.1 Pro Preview 43.5−18.7 n=6 Low
Gemini 3.5 Flash mentions only n=2 not enough data
Gemini 3.5 Flash Lite mentions only n=2 not enough data
Gemini 2.5 Pro mentions only n=1 not enough data
Gemini 3 Flash Preview mentions only n=1 not enough data
GLM 7 models
GLM 5.3 71.8+3.2 n=107 High
GLM 5.2 67.6−1.0 n=111 High
GLM 5.1 mentions only n=4 not enough data
GLM 5 mentions only n=3 not enough data
GLM 4.7 mentions only n=2 not enough data
GLM 4.5 Air mentions only n=1 not enough data
GLM 4.6V mentions only n=1 not enough data
Kimi 4 models
Kimi K3 62.5+2.4 n=156 High
Kimi K2.6 57.1−2.9 n=9 Low
Kimi K2.7 Code mentions only n=2 not enough data
Kimi K2.5 mentions only n=1 not enough data
Gemma 4 models
Gemma 4 31B 69.5+1.2 n=29 Medium
Gemma 4 26B A4B mentions only n=4 not enough data
Gemma 2 9B mentions only n=1 not enough data
Gemma 2 27B mentions only n=1 not enough data
MiniMax 4 models
MiniMax H3 61.4+4.3 n=36 High
MiniMax M3 48.4−8.7 n=14 Medium
MiniMax M2 mentions only n=1 not enough data
MiniMax M2.5 mentions only n=1 not enough data
Muse 2 models
Muse Glimmer 30B 56.3−11.5 n=15 Medium
Muse Spark 1.2 44.0−23.7 n=5 Low
MiMo 2 models
MiMo-V2.5 46.8−4.5 n=25 Medium
MiMo-V2.5-Pro mentions only n=3 not enough data

Scores use only comments that clearly identify the exact version; generic family mentions are tracked separately.