AI-reports @Razumasu on X

Tencent / model / guest in this test

Hy4 (full)

AngelSlim IQ2 GGUF, about 3.5 bits per weight, revision c65d2c9, running on Mac Studio M3 Ultra, 512 GB.

Guest, not a fair third contestant: different machine, different runtime, heavier compression.

12.2tok/sWriting speed, short prompt
17.5hoursTime for all 27 attempts
310GBFile size
Mac Studio M3 Ultra, 512 GB

Highlights

Hy4 full in Qwen3.8 Flash Next vs Hy4 Flash

A pelican riding a bicycle, drawn by Hy4 (full), attempt 1
From its drawingsPelican on a bicycle, attempt 1. See all three

A game it wrote

Game written by Hy4 (full), attempt 1, captured while runningHy4 full r1

These games were written for a keyboard and a big screen. On a phone they may not respond. They are best on a computer.

Test 01 / Pilot

Results in Qwen3.8 Flash Next vs Hy4 Flash

The full Hy4 is not a fair third contestant: different machine, different runtime, much heavier compression. Read its column as what the big model does on a Mac, not as a ranking against the other two.

Pelican on a bicycle

Opinion

A pelican riding a bicycle, drawn by Hy4 (full), attempt 1A pelican riding a bicycle, drawn by Hy4 (full), attempt 2A pelican riding a bicycle, drawn by Hy4 (full), attempt 3
Valid drawings, of 3 attempts
3

Octopus and penguin band

Opinion

An octopus conducting a brass band of penguins, drawn by Hy4 (full), attempt 1An octopus conducting a brass band of penguins, drawn by Hy4 (full), attempt 2An octopus conducting a brass band of penguins, drawn by Hy4 (full), attempt 3
Valid drawings, of 3 attempts
3

Playable browser game

Automatic load test; the rest is opinion

Complete games
2 of 3 (1 cut off at the answer limit)

Roast

Opinion

3.8 — not 3, not 4, 3.8. That's not a version, that's a Yelp rating for a dim sum place doing fine.

You didn't open-source a model, you opened a warehouse club.

My ranking (opinion)
Place 1 of 3

Text adventure

Rules check plus opinion

Complete games
0 of 3: empty answer on turn 2, probably a runtime fault, unconfirmed

Bug fix

Hidden tests, automatic

Bugs fixed of 3, per attempt
3, 3, 3
Minutes per attempt
over 60

Contract questions

Answer key

Score of 5, per attempt
5, 5, 5
Slip
None

12 steps with a contradiction

Answer key

Flagged the contradiction
2 of 3
Arithmetic right
3 of 3

Network design

Opinion

Words per plan
3,200 to 3,800
Access points proposed
6 to 11
Every attempt, with time and tokens
ChallengeAttemptStatusMinutesTokens
Playable browser game1Finished86.357,749
Playable browser game2Cut off at the answer limit90.160,000
Playable browser game3Finished35.524,737
Bug fix1Finished70.847,830
Bug fix2Finished78.452,692
Bug fix3Finished68.846,635
Network design1Finished66.445,093
Network design2Finished55.437,529
Network design3Finished65.844,274
Contract questions1Finished64.242,673
Contract questions2Finished31.421,004
Contract questions3Finished36.624,577
12 steps with a contradiction1Finished20.614,331
12 steps with a contradiction2Finished22.515,718
12 steps with a contradiction3Finished19.913,948
Pelican on a bicycle1Finished29.020,364
Pelican on a bicycle2Finished27.018,860
Pelican on a bicycle3Finished22.916,007
Octopus and penguin band1Finished31.121,474
Octopus and penguin band2Finished34.523,884
Octopus and penguin band3Finished27.719,745
Roast1Finished7.15,177
Roast2Finished5.94,313
Roast3Finished8.46,079
Text adventure1Empty answer12.28,665
Text adventure2Empty answer15.010,517
Text adventure3Empty answer15.610,949

Read the report Compare with the others

Build and settings

What exactly ran

Maker
Tencent
Build
AngelSlim IQ2 GGUF, about 3.5 bits per weight, revision c65d2c9
File size
310 GB
Machine
Mac Studio M3 Ultra, 512 GB
Runtime
llama.cpp PR #32517 at f8d4cc4, Metal
Sampling
temperature 0.9, top-p 1.0 (maker settings)
Thinking
on (default)

Speed

How fast it ran here

Writing speed, short prompt
12.2 tok/s
Writing speed after a 32K prompt
10.9 tok/s
Writing speed after a 127K prompt
10.0 tok/s
Reading speed, 32K prompt
79 tok/s
Reading speed, 127K prompt
75 tok/s
Wait for first token, 32K prompt
404 s
Wait for first token, 127K prompt
1,680 s (28 min)
Load time
13 s (warm)
Load time, cold
123 s (2.0 min)
Least memory left free
139 GiB
Time for all 27 attempts
17.5 hours
Tokens generated, all attempts
714,824 tokens
  • Short-prompt writing speed is the average of three runs; long-prompt rows are one run each.
  • Hy4 Flash speed rows come from the first pilot night (same files and runtime); Qwen rows from the rerun on the current build.
  • Speed tests use one fixed sampling setting for every model.
  • About nine tenths of what all three models wrote was thinking, not answer.