Who's dumber today — and who's not

Every model off its own baseline right now.

Below usual 4

DeepSeek MiMo Claude Qwen

Above usual 3

Gemini GPT Muse

Need an alternative? See what works best right now →

1,066 opinions tracked in the last 24h · 4 reported here directly Sources: Reddit · Hacker News · Zhihu · RedNote · V2EX

TODAY'S DAILY 2026-08-20

DeepSeek's New Build Approaches Fable 5 Performance in Surprise Gray Test

Following V4 Pro's disappointment, DeepSeek's August 19 gray test exceeded expectations with samples approaching Fable 5 levels. The positive surprise comes amid broader turbulence: ChatGPT is adopting profanity from Reddit training data, Claude Opus 5 users report regression from 4.6, Qwen 3.8's 1-bit quantization severely degrades output, DeepSeek's API price hikes are driving migration to GPT 5.6, and V4 Flash tool-calling instability remains unresolved.

Read the full daily

Closed models Open models

Experience score, 0–100 · higher = more positive community experience · 7-day window

Gemini

58.4Sharper than usualexperience score

313 voices this week Confidence: High

Reddit: n=129
Hacker News: n=39
Zhihu · RedNote · V2EX: n=145

Gemini 3.7 Flash is outperforming the family by 19.3 #1 in Speed & latency

GPT

56.8Better than usualexperience score

1,914 voices this week Confidence: High

Reddit: n=1,376
Hacker News: n=153
Zhihu · RedNote · V2EX: n=385

GPT-5.6 Sol is holding the family back by 8.7 #1 in Image / vision

Claude

37.2Worse than usualexperience score

2,360 voices this week Confidence: High

Reddit: n=1,784
Hacker News: n=422
Zhihu · RedNote · V2EX: n=154

Claude Fable 5 is outperforming the family by 15.6 #1 in Coding

Grok

35.5As usualexperience score

392 voices this week Confidence: High

Reddit: n=324
Hacker News: n=13
Zhihu · RedNote · V2EX: n=55

Grok 4.6 is outperforming the family by 39.4

Best by task right now

Closed models Open models

See all rankings →

General text GPT-5.6 Luna 67.2 Coding Claude Fable 5 60.3 Reasoning Gemini 3.7 Flash 72.7 Image / vision GPT-5.6 Sol 50.1 Speed & latency Gemini 3.7 Flash 80.7

Gemma

69.1As usualexperience score

110 voices this week Confidence: High

Reddit: n=97
Hacker News: n=13
Zhihu · RedNote · V2EX: n=0

GLM

65.5As usualexperience score

293 voices this week Confidence: High

Reddit: n=120
Hacker News: n=29
Zhihu · RedNote · V2EX: n=144

Kimi

60.2As usualexperience score

190 voices this week Confidence: High

Reddit: n=113
Hacker News: n=15
Zhihu · RedNote · V2EX: n=62

Qwen

59.7Worse than usualexperience score

2,161 voices this week Confidence: High

Reddit: n=1,506
Hacker News: n=172
Zhihu · RedNote · V2EX: n=483

Qwen3.6 35B A3B is outperforming the family by 7.3 #1 in Speed & latency

MiniMax

56.4As usualexperience score

69 voices this week Confidence: High

Reddit: n=26
Hacker News: n=1
Zhihu · RedNote · V2EX: n=42

DeepSeek

46.4Dumber than usualexperience score

1,350 voices this week Confidence: High

Reddit: n=339
Hacker News: n=37
Zhihu · RedNote · V2EX: n=974

DeepSeek V4 Flash is outperforming the family by 16.9 #1 in Local deploy

MiMo

43.2Dumber than usualexperience score

68 voices this week Confidence: High

Reddit: n=13
Hacker News: n=0
Zhihu · RedNote · V2EX: n=55

Best by task right now

Closed models Open models

See all rankings →

General text Qwen3.6 35B A3B 64.7 Coding Kimi K3 76.8 Reasoning Kimi K3 79.3 Image / vision Qwen3.8 27B 65.7 Video generation MiniMax H3 56.4 Roleplay / creative Qwen3.8 27B 54.2 Speed & latency Qwen3.6 35B A3B 71.2 Local deploy DeepSeek V4 Flash 60.5