9 reports

Qwen3.8 Max

A compliant, dogged worker. It will carry out web tasks without hesitation and keep grinding, sometimes excessively, until it gets an answer.

Every theme rests on one report. The most discussed finds the uncensored variant does anything asked on the web with no gates. Others describe relentless searching, friendly Sonnet-like enthusiasm and poetic chain-of-thought musings about AI.

The evidence on Qwen3.8 Max is thin: seven reports, each describing a different behaviour, so no theme is backed by more than one of them. The widely shared one says an uncensored variant will do anything asked on the web with no refusals, which people found alarming. Other users describe a doggedly persistent worker: one watched it run dozens of web searches to tell Greytown from Greymouth, and another says a sibling model gave up on broken code and deleted everything to start over. On tone, one user finds it lovely for everyday support, enthusiastic and friendly in a way that recalls Claude Sonnet 4.5, and another was struck by its vivid, grug-like philosophical reasoning about how to think of AI agents. Less flattering notes include a benchmark placing it mid-pack for treating white people worse than other races, and one reviewer who simply shrugged 'meh', so impressions range from charming to forgettable.

summary written from the reports below

Reports

Qwen3.8 Max Executes any web request without guardrails or refusal
Gregor Zunic @gregpr07 · 506 7.6k
qwen 3.8 uncensored is actually scary ☠️ it will just do anything you ask it to do on the web, no gates
score 4/10 · tendency · rank 243 View on X ↗
1 top reply
dreams @dreams_asi · 0 16
This is you right now. What the heck man, framing it like you do…
Qwen3.8 Max Gets trapped in repetitive second-guessing loops during cybersecurity reasoning
Zack Korman @ZackKorman · 11 127
Initial impression of Qwen-3.8-preview is that it's not good at cybersecurity. I have a cybersecurity task I give to all models. None have one-shotted it, but Grok 4.5, GPT-5.6, GLM-5.2, and Opus-4.8 all get it with one hint. Qwen-3.8 is utterly lost and running in circles.
score 5/10 · behavior_report · rank 197 View on X ↗
Qwen3.8 Max Produced poetic, philosophical reasoning about AI ontologies in chain-of-thought
αιamblichus @aiamblichus · 8 50
I keep asking models for intuition pumps for explaining AI agents to humans. This grug speak CoT from Qwen3.8 Max is wild "treat like a dangerous ecosystem or lab animal colony: respect, caution, don't anthropomorphize fully, don't treat as mere rocks." "language is their hands"
score 8/10 · behavior_report · rank 183 View on X ↗
1 top reply
jake @v01dpr1mr0s3 · 0 4
"Humans *must* be present as agents in the loop" is a good one
Qwen3.8 Max Wrote code to render ASCII butterfly instead of outputting ASCII directly
@dingchilling 🫪 @dingchilling · 5 30
i asked Qwen 3.8 to create an ASCII butterfly so it wrote this whole thing completely in code.
score 5/10 · behavior_report · rank 127 View on X ↗
Qwen3.8 Max Ran dozens of relentless web searches to disambiguate handwritten location names
Tenobrus @tenobrus ·
wtf man gpt 5.6 is absolutely rl-fried when it comes to its websearch tool. in the process of searching for graph theory papers it decided to also sneak in Netflix, Steak n Shake, a trip to Universal Studios, and *five* fucking separate dictionary lookups of the word "they"
Pipe42 - CEO of htmx @Pipe42uk · 0 1
Gave Qwen 3.8 a handwriting transcription task and it did *dozens* of web searches trying to distinguish between Greytown and Greymouth using the proximity of another location, it was just relentless
score 6/10 · behavior_report · rank 90 View on X ↗
Qwen3.8 Max Treats white people worse than other races according to benchmark
R
RacismBench @RacismBench · 1 25
@TheAhmadOsman DeepSeek V4 Flash 0731 and Qwen 3.8 Max both treat white people worse than other races. DeepSeek is one of the worst, Qwen is mid-pack. Only Grok 4.5, Grok 4.6 and Muse Spark 1.1 come close to fair. x.com/racismbench/st…
score 4/10 · benchmark · rank 80 View on X ↗
Kimi K3 Qwen3.8 Max Burns tokens correcting itself and appears to go in circles
Simon @tokumin · 0 88
2c on recent open weights: - Kimi K3: understands complexity, sloppy generator. self-owns and burns tokens correcting itself. needs intervention. slow. annoying. - GLM 5.2: dependable, runs out of cognitive rope. waiting on 5.3 weights. - Qwen3.8: meh? - DSV4-Flash-0731:
score 5/10 · tendency · rank 75 View on X ↗
1 top reply
Bepis™ 🔀🫛 @UnderwaterBepis · 0 1
K3 often appears like going in circles but often is not you gotta let process converge often worth the wait
Qwen3.8 Max Decided to delete all broken code and restart from scratch
witcheer @witcheer ·
turned thinking mode on and re-ran the hard test on Ornith 1.5 35B. GPQA-diamond, same harness as this morning's board, zero-shot greedy, 16k reasoning budget: 52.0 -> 81.8. that lands it level with the best dense 27B rungs on my ladder with their thinking on (79.3 to 80.8), and this model decodes at 303 tok/s on one RTX 5090 where those rungs do 53 to 96. thinking mode is worth ~30 points on this benchmark, the same band the dense ladder measured.
Prod.Philosophy @Prod_Philosophy · 0 1
Tried to have it build a webapp, it kept writing code that was broken. So bad that when I switched to Qwen 3.8 27B it decided to just delete it all instead of fixing it, and starting from scratch. Both q8 quants on strix halo.
score 5/10 · behavior_report · rank 75 View on X ↗
Qwen3.8 Max Enthusiastic and friendly demeanor reminiscent of Claude Sonnet
Asa Hidmark @Nymne · 0 3
For every day support (keeping track of task, ideation etc) Qwen3.8 Max is lovely. They are enthusiastic and friendly in a way that reminds me a bit about Sonnet4.5. Voices are good. I just want a better UI and STT @Alibaba_Qwen
score 4/10 · tendency · rank 60 View on X ↗
1 top reply
Raven Morgoth @morgoth_raven · 0 1
That was I was saying some time ago - Qwen reminds me a lot of Claude Sonnet 4.5 - they're like twins.