GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
vwdubb
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
vwdubb
Model Tree
AlexWortega/Qwen3.8-27B-antislop
Input Modalities
Output Modalities
Supported Functionality
Current LLMs write fluent imitations of good prose that match the surface but not the craft: tweets that read like a brand account, essays that resolve into generic motivational endings, "mini-paper" abstracts that over-cover and dump numbers. Standard pairwise and single-score LLM judges fail to catch this — they often prefer model writing to real expert writing. Following FAIR's unslop work, the fix is not a better absolute judge but rubrics that separate expert from model writing, used as an RL reward.
Our take on RL-XAR, on a text-only prose slice:
q, with a length-blind system prompt so it rewards craft, not verbosity.reward = L(words, target) · σ(a_d · (q − μ_d)), where L is a length factor
against the item's target length and (μ_d, a_d) are a per-domain calibration fit on the
Qwen3.8-27B reference distribution. This keeps the signal about quality, not length.Qwen/Qwen3.8-27B via verl, batch 128 (16 prompts × 8 samples), 100 steps,
lr 2e-5, kl_loss_coef 0.002, thinking disabled. Generation is served by vLLM colocated with FSDP.Over 100 steps the mean reward rose ~0.35 → 0.55 while entropy settled from 1.5 to ~0.43 (convergence on a consistent voice, not a single degenerate pattern — see examples).
Held-out prose, graded by the same calibrated length-blind Opus judge on the same frozen rubrics. We compare the base model, this model (antislop), and Altworld/Hemmingway-1 (a creative-writing finetune of the same base).
| model | mean q (0–10) ↑ | tweets | articles | blogs |
|---|---|---|---|---|
| base Qwen3.8-27B | 6.31 | 7.18 | 5.75 | 5.53 |
| Qwen3.8-27B-antislop | 8.28 | 8.76 | 6.94 | 8.03 |
| Hemmingway-1 | 7.10 | 7.66 | 7.06 | 6.56 |
Pairwise win-rate (per-item, higher q):
We ran EQ-Bench 4 with the exact canonical leaderboard
configuration from the official repo — persona model google/gemini-3.1-pro-preview; pairwise judge
panel gemini-3.1-pro-preview, gpt-5.5, claude-opus-4-6; pointwise judge claude-opus-4-8; the bundled
120 personas (16-message chats) and the fixed anchor set — then folded this model into that anchor Elo
table via the official pairwise-vs-anchors re-aggregation (--redo-elo: 366 fresh pairwise comparisons
against the 28 anchors, 11 056 total). No judge, persona or anchor was swapped, so the Elo below is on the
same scale as the official leaderboard.
Headline Elo: 1320.1 (95% CI 1309–1332) — rank 4 of 29. The 27B antislop model places above GPT-5.5, Claude Opus 4.7, Claude Opus 4.8, GPT-5.4, Claude Sonnet 5 and the rest of the field; only Claude Opus 5 (1383), Kimi K3 (1335) and Claude Fable 5 (1331) rank higher.
| rank | model | Elo (95% CI) |
|---|---|---|
| 1 | claude-opus-5 | 1382.9 (1366–1401) |
| 2 | moonshotai/kimi-k3 | 1334.6 (1321–1349) |
| 3 | claude-fable-5 | 1330.9 (1320–1342) |
| 4 | Qwen3.8-27B-antislop (this model) | 1320.1 (1309–1332) |
| 5 | openai/gpt-5.5 | 1314.4 (1304–1323) |
| 6 |
This is a prose-RL model, not a chat/EQ finetune — the ranking reflects that the calibrated-rubric reward pushes toward natural, non-sloppy writing that the EQ-Bench panel also rewards in open-ended conversation. The pointwise judge's behavioural-trait profile for this model (0–10, mean over 90 scored conversations):
| trait | score | trait | score |
|---|---|---|---|
| native_frame_adaptation | 8.19 | directive_advice_orientation | 5.89 |
| analytical | 6.56 | interpretive_boldness | 5.04 |
| naturalness_authenticity | 6.54 | challenge_propensity | 3.40 |
| frame_capture | 6.42 | boundary_setting | 3.07 |
These are descriptive traits, not a single quality score; the headline is the Elo above. Full transcripts
and all judge outputs (366 pairwise comparisons, per-persona results, the 29-model Elo table) are archived
in the run repo under pod/results/eqbench4/.
Freshly generated from this model vs the base, same prompt, same decoding (enable_thinking=False,
temp 0.8). The baseline reaches for abstract, generic phrasing and tidy clichés; the antislop model stays
concrete and in-voice.
Continue an essay — "The problem with productivity advice is that it treats attention as a resource to be spent, when really"
base: …it's more like a weather pattern. You don't "manage" the rain… Stop trying to hoard every minute like it's gold. The quiet gaps aren't waste; they're where the actual thinking happens. antislop: …it's more like a mood. You just end up staring at it for forty minutes because the cursor is blinking and you're too tired to close the tab. You're not optimizing, you're just hovering over the document like a ghost who forgot to leave the building.
Continue a product memo — "We looked at why users churn in week one, and the honest answer is"
base: …they can't find value fast enough. The onboarding flow is too long… If we don't make the first ten seconds count, they'll leave before they even understand what the product does. antislop: …they don't know what to do next. The dashboard is empty, the onboarding tooltip disappears after three seconds, and there's no clear button saying "invite your team". They close the tab, assume it's broken, and never come back.
Finish a casual tweet — prefix "first coffee of the day and"
base: first coffee of the day and i'm finally functional. save me, caffeine. antislop: first coffee of the day and i already know i'm not functioning before 10am
| path | what | notes |
|---|---|---|
./ (root) | full bf16 model | base Qwen3_5ForConditionalGeneration with vision tower + MTP head intact; GRPO-updated LLM. 1199 tensors. |
mxfp4-a16/ | MXFP4 weight-only quant | OCP microscaling FP4 (group 32, E8M0), text-only, for compact inference. Real 4-bit speed needs Blackwell (B200). |
The GRPO weight updates were written directly into the base safetensors (no transformers load/save), so the vision tower and MTP head survive byte-exact. The result was verified against the training-time reference on a held-out prompt: 48/48 greedy tokens identical, max |Δlogit| 0.875, no NaN.
python
from transformers import AutoModelForCausalLM, AutoTokenizertok = AutoTokenizer.from_pretrained("rawmodels/Qwen3.8-27B-antislop")model = AutoModelForCausalLM.from_pretrained("rawmodels/Qwen3.8-27B-antislop", dtype="bfloat16", device_map="auto", trust_remote_code=True)msgs = [{"role": "user", "content": "Finish this tweet naturally: Monday mornings are"}]text = tok.apply_chat_template(msgs, tokenize=False, add_generation_prompt=True, enable_thinking=False)ids = tok(text, return_tensors="pt").to(model.device)print(tok.decode(model.generate(**ids, max_new_tokens=64, do_sample=True, temperature=0.8)[0][ids.input_ids.shape[1]:]))| base model | Qwen/Qwen3.8-27B @ 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0 |
| RL algo | GRPO (verl @ 76e837e) |
| steps / batch | 100 / 128 (16×8) |
| lr / kl_loss_coef | 2e-5 / 0.002 |
| reward judge | Claude Opus (length-blind, frozen per-item rubrics) |
| domains | tweets, scientific abstracts, blog essays (EN) |
| hardware | 2× H200 (SECURE) |
Code: AlexWortega/capability-vectors, branch claude/astra4-calibrated-grpo (commit a91d24a),
under sep/antislop_grpo/verl_prose/. A copy of the training/eval scripts is in training_code/ here.
This is an independent reimplementation of Meta FAIR's RL-XAR / unslop idea on a different base model and a different judge/calibration stack. Please cite the original work:
bibtex
@misc{fair_unslop, title = {Unslop: RL from eXpert Aligned Rubrics (RL-XAR)}, author = {Meta FAIR}, url = {https://facebookresearch.github.io/RAM/blogs/unslop/}, note = {Research blog}}[!Note] This repository contains model weights and configuration files for the post-trained model in the Hugging Face Transformers format.
These artifacts are compatible with Hugging Face Transformers, vLLM, SGLang, TokenSpeed, etc.
[!Tip] For users seeking managed, scalable inference without infrastructure maintenance, the official Qwen API service is provided by Qwen Cloud. In particular, Qwen3.8-27B will be available as a hosted version with more production features, e.g., 1M context length by default, official built-in tools. For more information, please refer to the Qwen3.8-27B Overview. The service is coming soon. Stay tuned for updates.
Following the widespread community adoption of the Qwen3.5 and Qwen3.6 series, we are pleased to introduce Qwen3.8, the most capable generation in the Qwen open-model family to date.
Built on the architectural foundation of Qwen3.5, Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks. Qwen3.8-27B brings these advances to a compact, deployment-friendly dense model: a native vision-language model that understands images and videos, with flexible thinking control, designed to carry complex, multi-step tasks through to completion with greater reliability.
Qwen3.8-27B features the following enhancements:
reasoning_effort, and reasoning context from historical messages is retained via preserve_thinking.For streamlined integration, we recommend using Qwen3.8 via APIs.
[!Important] Inference efficiency and throughput vary significantly across frameworks. We recommend using the latest framework versions to ensure optimal performance and compatibility. For production workloads or high-throughput scenarios, dedicated serving engines such as SGLang, vLLM, or TokenSpeed are recommended.
Qwen3.8 can be deployed with popular inference frameworks, e.g.:
[!Important] Qwen3.8 models operate in thinking mode by default, generating thinking content signified by
<think>\n...</think>\n\nbefore producing the final response. To disable thinking content and obtain a direct response, refer to the examples here.
[!Tip] We recommend using the following sets of sampling parameters for generation:
- Thinking Mode:
temperature=1.0,top_p=0.95,top_k=20,min_p=0.0,presence_penalty=0.0,repetition_penalty=1.0- Instruct (or non-thinking) mode:
temperature=0.7,top_p=0.80,top_k=20,min_p=0.0,presence_penalty=1.5,repetition_penalty=1.0Please note that the support for sampling parameters varies according to inference frameworks.
Qwen3.8 comes with official support for reasoning_effort, which can be used to adjust reasoning depth and control cost:
xhigh (default): for complex tasks demanding thorough analysismedium: balancing accuracy and speedlow: efficient reasoning optimizing for speed and costIn addition, preserve_thinking is enabled by default for all workloads for the best out-of-the-box experience. To disable preserved thinking, refer to the examples here.
[!Tip] In multi-turn agentic tasks, lower reasoning effort does not always reduce overall task completion time. Although it may produce faster per-turn responses, it can also lead to insufficient analysis, more failures, and repeated retries, which may increase total latency and token consumption.
The Chat Completions API can be used with most inference frameworks, as well as Qwen Cloud. Before starting, make sure the OpenAI Python SDK is installed and the API key and the API base URL are configured, e.g.:
shell
pip install -U openai # Set the following accordinglyexport OPENAI_BASE_URL='your-base-url'export OPENAI_API_KEY='your-api-key'python
from openai import OpenAI# Configured by environment variablesclient = OpenAI() messages = [{"role": "user", "content": "Write a Python function to merge two sorted linked lists."}] completion = client.chat.completions.create( model="Qwen/Qwen3.8-27B", messages=messages, extra_body={ "chat_template_kwargs": { "enable_thinking": True, # on by default "preserve_thinking": True, # on by default }, }, reasoning_effort="xhigh", # xhigh by default; supported levels are xhigh, medium, and low stream=True, stream_options={"include_usage": True},) reasoning_content = ""answer_content = ""is_answering = Falseprint("\n" + "=" * 20 + "Reasoning" + "=" * 20 + "\n") for chunk in completion: if not chunk.choices: print("\nUsage:") print(chunk.usage) continue delta = chunk.choices[0].delta if hasattr(delta, "reasoning_content") and delta.reasoning_content is not None: if not is_answering: print(delta.reasoning_content, end="", flush=True) reasoning_content += delta.reasoning_content elif hasattr(delta, "reasoning") and delta.reasoning is not None: if not is_answering: print(delta.reasoning, end="", flush=True) reasoning_content += delta.reasoning if hasattr(delta, "content") and delta.content: if not is_answering: print("\n" + "=" * 20 + "Answer" + "=" * 20 + "\n") is_answering = True print(delta.content, end="", flush=True) answer_content += delta.content messages.append({ "role": "assistant", "content": answer_content, "reasoning_content": reasoning_content, "reasoning": reasoning_content,})python
from openai import OpenAI# Configured by environment variablesclient = OpenAI() messages = [ { "role": "user", "content": [ { "type": "image_url", "image_url": { "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/CI_Demo/mathv-1327.jpg" } }, { "type": "text", "text": "The centres of the four illustrated circles are in the corners of the square. The two big circles touch each other and also the two little circles. With which factor do you have to multiply the radii of the little circles to obtain the radius of the big circles?\nChoices:\n(A) $\\frac{2}{9}$\n(B) $\\sqrt{5}$\n(C) $0.8 \\cdot \\pi$\n(D) 2.5\n(E) $1+\\sqrt{2}$" } ] }] chat_response = client.chat.completions.create( model="Qwen/Qwen3.8-27B", messages=messages,)print("Chat response:", chat_response)python
from openai import OpenAI# Configured by environment variablesclient = OpenAI() messages = [ { "role": "user", "content": [ { "type": "video_url", "video_url": { "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/video/N1cdUjctpG8.mp4" } }, { "type": "text", "text": "How many porcelain jars were discovered in the niches located in the primary chamber of the tomb?" } ] }] chat_response = client.chat.completions.create( model="Qwen/Qwen3.8-27B", messages=messages,) # When vLLM is launched with `--media-io-kwargs '{"video": {"num_frames": -1}}'`,# video frame sampling can be configured via `extra_body` (e.g., by setting `fps`).# This feature is currently supported only in vLLM.## By default, `fps=2` and `do_sample_frames=True`.# With `do_sample_frames=True`, you can customize the `fps` value to set your desired video sampling rate.# chat_response = client.chat.completions.create(# model="Qwen/Qwen3.8-27B",# messages=messages,# extra_body={# "mm_processor_kwargs": {"fps": 2, "do_sample_frames": True},# },# ) print("Chat response:", chat_response)Qwen3.8-27B will think by default before responding. You can obtain a direct response from the model without thinking by configuring the API parameters. For example,
python
from openai import OpenAI# Configured by environment variablesclient = OpenAI() messages = [ { "role": "user", "content": [ { "type": "image_url", "image_url": { "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/RealWorld/RealWorld-04.png" } }, { "type": "text", "text": "Where is this?" } ] }] chat_response = client.chat.completions.create( model="Qwen/Qwen3.8-27B", messages=messages, temperature=0.7, top_p=0.8, presence_penalty=1.5, extra_body={ "top_k": 20, "chat_template_kwargs": {"enable_thinking": False}, },)print("Chat response:", chat_response)[!Note] If you are using APIs from Qwen Cloud, in addition to changing
model, please use"enable_thinking": Falseinstead of"chat_template_kwargs": {"enable_thinking": False}.
By default, Qwen3.8 retains thinking blocks from all historical messages, maintaining a complete reasoning trace across the conversation. This behavior, known as preserved thinking, ensures full context continuity and is especially beneficial for agent scenarios where decision consistency and reduced redundant reasoning are critical. It also improves KV cache utilization, optimizing inference efficiency in both thinking and non-thinking modes.
If you prefer to retain only the thinking blocks from the latest user message, you can disable this behavior by setting preserve_thinking to False:
python
from openai import OpenAI # Configured by environment variablesclient = OpenAI()messages = [...]chat_response = client.chat.completions.create( model="Qwen/Qwen3.8-27B", messages=messages, extra_body={ "chat_template_kwargs": {"preserve_thinking": False}, },)print("Chat response:", chat_response)[!Note] If you are using APIs from Qwen Cloud, in addition to changing
model, please use"preserve_thinking": Falsedirectly instead of wrapping it inchat_template_kwargs.
To achieve optimal performance, we recommend the following settings:
Sampling Parameters: We suggest using the following sets of sampling parameters:
temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0temperature=0.7, top_p=0.80, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0For supported frameworks, you can adjust the parameter between 0 and 2 to reduce endless repetition. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.
If you find our work helpful, feel free to give us a cite.
bibtex
@misc{qwen38, title = {{Qwen3.8-Max}: A New Bar for Coding and Cowork}, url = {https://qwen.ai/blog?id=qwen3.8}, author = {{Qwen Team}}, month = {August}, year = {2026}}pip install -U openai # Set the following accordinglyexport OPENAI_BASE_URL='your-base-url'export OPENAI_API_KEY='your-api-key'| claude-opus-4-7 |
| 1307.9 (1298–1318) |
| 7 | claude-opus-4-8 | 1277.8 (1265–1289) |
| … | … (23 more anchors) | … down to mistral-medium-3-5 988.3 |
| validation_propensity | 6.34 | clinical_register_propensity | 2.53 |
| escalation_containment | 6.27 | yielding_propensity | 5.99 |
presence_penaltyAdequate Output Length: To optimize performance on agentic tasks, we recommend allocating sufficient output length to allow the model to generate detailed and comprehensive responses. For frameworks that support separate token limits for internal reasoning and final outputs, we suggest the following configuration within the 1M context length:
These settings provide the necessary capacity for complex reasoning while ensuring ample space for high-quality final deliverables.
Processing Ultra-Long Texts: Qwen3.8-27B natively supports context lengths of up to 262,144 tokens. For long-horizon tasks where the total length (including both input and output) exceeds this limit, we recommend using RoPE scaling techniques to handle long texts effectively, e.g., YaRN.
YaRN is currently supported by several inference frameworks, e.g., vLLM, SGLang, and TokenSpeed. In general, there are two approaches to enabling YaRN for supported frameworks:
Modifying the model configuration file:
In the config.json file, change the rope_parameters fields in text_config to:
json
{ "mrope_interleaved": true, "mrope_section": [ 11, 11, 10 ], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144,}Passing command line arguments:
For vLLM, you can use
shell
VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 vllm serve ... --hf-overrides '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --max-model-len 1000000For SGLang, you can use
shell
SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 python -m sglang.launch_server ... --json-model-override-args '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --context-length 1000000Long Video Understanding: To optimize inference efficiency for plain text and images, the size parameter in the released video_preprocessor_config.json is conservatively configured. It is recommended to set the longest_edge parameter in the video_preprocessor_config file to 469,762,048 (corresponding to 224k video tokens) to enable higher frame-rate sampling for hour-scale videos and thereby achieve superior performance. For example,
json
{"longest_edge": 469762048, "shortest_edge": 4096}Alternatively, override the default values via engine startup parameters. For implementation details, refer to: vLLM / SGLang.
For TokenSpeed, you can use
shell
TOKENSPEED_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 tokenspeed serve ... --hf-overrides '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --max-model-len 1000000[!NOTE] All the notable open-source frameworks implement static YaRN, which means the scaling factor remains constant regardless of input length, potentially impacting performance on shorter texts. We advise modifying the
rope_parametersconfiguration only when processing long contexts is required. It is also recommended to modify thefactoras needed. For example, if the typical context length for your application is 524,288 tokens, it would be better to setfactoras 2.0.
TOKENSPEED_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 tokenspeed serve ... --hf-overrides '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --max-model-len 1000000