In this piece
Open-Weight LLMs 2026: DeepSeek, Qwen, Kimi, GLM, Llama and Muse
Start workstation evaluations with Qwen3.8-27B and Muse Glimmer-30B. For cluster deployments, compare DeepSeek V4.1 Flash-1017, Kimi K3.1 Pro and the other larger checkpoints under their own licenses.
"Open weight" only says that weights are downloadable. It does not guarantee an Open Source Initiative-approved license, low hardware cost, identical hosted and self-hosted behavior, or usable quality at the advertised maximum context. Shortlist by the exact checkpoint and license, then test your own tasks.
This buyer shortlist was reviewed on 11 November 2026. Check each model at your expected context length and concurrency. Provider benchmark scores reflect the tools and settings used in each evaluation.
Need a model shortlist that fits your tasks, EU boundary, and GPU budget?
Shortlist Models for My WorkloadWhat separates the current open-weight families?
- DeepSeek. V4.1 Flash-1017 accepts text and images and produces text. V4 Pro-0813 remains a text-only option. Both releases carry MIT terms and document a 1M-token context. DeepSeek’s hosted deepseek-flash API name now serves the October Flash checkpoint.
- Qwen3.8. Alibaba’s official Qwen3.8 collection includes a 27B multimodal model and a 2.4T total, 95B-active sparse model. The 27B card states a native 262,144-token window, extension to 1M with YaRN, and Apache 2.0 licensing. The 2.4T checkpoint uses the Qwen3.8-Max License, including conditions for very large products and large MaaS or AI work-assistant businesses. Extension is a serving configuration, not evidence that every task remains reliable at 1M.
- Kimi K3.1 Pro. Moonshot’s Kimi K3.1 Pro card lists 2.8T total and 104B active parameters, text and image input and a 1M-token context. The Kimi K3.1 Pro License sets separate conditions for large model-as-a-service businesses and very large commercial products. The extra conditions exempt qualifying internal use and access via Moonshot’s official products or certified inference partners.
- GLM-5.3. Z.ai released GLM-5.3 after GLM-5.2. It uses the same 753B base model, supports 1M-token evaluation, and changes the post-training for coding and long-horizon work. Its GLM-5.3 License is permissive, but it is not labelled MIT, so the earlier MIT claim is no longer accurate for the current checkpoint.
- Llama 4. Meta’s model card lists Scout at 109B total and 17B active parameters with 10M context, and Maverick at 400B total and 17B active with 1M context. The use policy withholds the multimodal license grant from EU-domiciled individuals and companies whose principal place of business is in the EU. It expressly says that this restriction does not apply to end users of a product incorporating the models.
- Muse Glimmer-30B. Meta’s official model card identifies an August 2026 release under Apache 2.0, approximately 29.6B total parameters including its perception encoder, text and image input, text output and a published 131,072+ context. It targets local agent workflows. This is a separate model and license from Llama 4, not a newly removed restriction on Llama itself.
Two more cluster candidates are GLM-5.3-Flash, a natively multimodal model with 320B total and 18B active parameters under MIT, and Qwen3.8-Flash-Next, a multimodal release under Qwen Community License 1.0. Meta’s October Muse Spark Open uses 816B total and 42B active parameters and a 1M-token context. Its gated Muse Model License includes attribution, use-policy and model-as-a-service conditions.
Tencent’s Hy4 preview supplies downloadable Apache 2.0 weights. The 4 November Hy4 launch added visual understanding to the hosted release and announced that its weights would follow. Tencent Cloud advises moving from hy4-preview to hy4 before 7 December 2026 Beijing time. Use the published preview checkpoint for self-hosted tests and confirm the exact final weights and terms separately.
Germany’s Soofi S sits in a different category: a sovereign German-English project. Our Soofi S buyer’s review separates public evidence from release plans. Our Kimi K3 API review for EU companies treats Moonshot’s hosted service as a separate procurement decision from running the downloadable weights.
Size, context, license, and deployment
| Family | Representative current checkpoint | Published size | Published context | Weight license | Deployment implication |
|---|---|---|---|---|---|
| DeepSeek | V4.1 Flash-1017 / V4 Pro-0813 | 763B / 1.7T (Safetensors counts) | 1M | MIT | Large multi-GPU deployment; calculate from the exact weight format |
| Qwen | Qwen3.8-27B / 2.4T-A95B | 27B dense / 2.4T total, 95B active | 27B: 262,144 native, extensible to 1M | Apache 2.0 / Qwen3.8-Max License | 27B is materially easier to host; the 2.4T tier is cluster-scale |
| Kimi | Kimi K3.1 Pro | 2.8T total / 104B active | 1M | Kimi K3.1 Pro License | Cluster-scale checkpoint plus model-specific license review |
| GLM | GLM-5.3 | 753B total | 1M | GLM-5.3 License | Serious multi-GPU or offloaded experimental serving |
| Llama | Llama 4 Scout / Maverick | 109B / 17B active; 400B / 17B active | 10M / 1M | Llama 4 Community License | Resolve the EU multimodal developer restriction first |
| Muse | Muse Glimmer-30B | Approximately 29.6B total, dense | 131,072+ | Apache 2.0 | Local-agent candidate; size the quantization, encoder and KV cache together |
These are maximum context claims, not quality guarantees. Long sequences increase KV-cache demand and can reduce concurrency. Quantization changes memory needs and sometimes output quality. Do not convert parameter counts into a fixed GPU count without specifying precision, runtime, context, batch size, and redundancy.
Separate workstation candidates from cluster deployments
Qwen3.8-27B and Muse Glimmer-30B belong in a different deployment conversation from trillion-parameter checkpoints. Meta gives 24 GB and 32 GB target configurations for its quantized Muse packages, and 64 GB for full precision. Those are vendor targets for specified configurations, not a promise that every runtime, context length or concurrent workload will fit. Preserve headroom for KV cache, images, any drafter and the rest of the application. Our Muse Glimmer local-agent guide covers the specific packaging and tests rather than duplicating that review here.
Which family should you test for coding and agents?
The primary sources support a shortlist, not a universal ranking. DeepSeek’s V4.1 Flash-1017 card, Qwen3.8's model card, Kimi K3.1 Pro’s card, and GLM-5.3's card all publish coding and agent benchmarks, but they use different agent harnesses, reasoning settings, task versions, and sometimes internal test sets. Compare numbers only where the task, split, harness, tool access, and scoring method match.
- For a workstation-scale multimodal agent: evaluate Qwen3.8-27B and Muse Glimmer-30B side by side. Both use Apache 2.0 for these exact checkpoints, but context, tool-use behavior, quantization and runtime support differ.
- For large coding or tool-use evaluations: test DeepSeek V4.1 Flash-1017, GLM-5.3, and Kimi K3.1 Pro on completed-task rate, recovery behavior, latency, and total serving cost.
- For very long context: test retrieval accuracy and task completion at your actual lengths. A 1M or 10M limit only establishes accepted input size.
- For Llama 4 in the EU: obtain legal review before a developer downloads or deploys the multimodal weights. Do not misread the end-user exception as a developer license.

"A model card tells you what to test and under which license. Your own eval tells you what to ship. Keep those two decisions separate."
Can you self-host these in the EU?
DeepSeek V4, Qwen3.8-27B and Muse Glimmer-30B use standard permissive licenses for the named weights. Qwen3.8-2.4T-A95B, GLM-5.3 and Kimi K3.1 Pro use model-specific texts that require separate review. Muse Spark Open has its own terms, including attribution and use-policy conditions. Llama 4’s multimodal policy withholds developer rights for the specified EU-based users while excepting end users of an incorporating product. That Llama-specific restriction is not Muse’s license. No weight license by itself establishes GDPR compliance.
Self-hosting can keep inference inside an EU-controlled environment, but it does not automatically establish GDPR compliance. You still need a lawful basis, data minimization, retention rules, access controls, processor and transfer analysis where relevant, logging policy, and an incident process. A hosted endpoint must be assessed separately from the downloadable checkpoint because its data location and contract are service-specific.
A defensible selection process
- Name the exact checkpoint. Family labels hide material differences in weights, modalities, context configuration, and license.
- Clear license and jurisdiction. Record the license revision and have counsel review model-specific restrictions when the deployment is consequential.
- Budget the actual serving configuration. Include precision, context, concurrency, runtime overhead, redundancy, and network topology.
- Run the same private eval. Use representative tasks, stable prompts, identical tool access, and explicit pass conditions across at least two candidates.
- Pin and monitor. Record the weight revision, tokenizer, runtime, quantization, and safety configuration. Re-evaluate before changing any of them.
Why the eval harness matters
Public benchmarks can suffer from contamination, overfitting, and harness effects. More importantly, they are not your workload. A few dozen representative tasks with clear pass criteria can expose failure modes that an aggregate leaderboard hides. Measure completed-task quality, latency, cost, tool errors, long-context retrieval, and recovery after a failed step.
Frequently Asked Questions
Does Meta’s Llama 4 EU restriction apply to Muse Glimmer?
Which models should a small team test locally first?
Is the longest advertised context the best model?
Are these Wavect benchmark results?
Final thoughts
The November 2026 shortlist distinguishes workstation-scale Qwen3.8-27B and Muse Glimmer-30B from the much larger checkpoints. Keep license decisions checkpoint-specific, particularly for Muse versus Llama 4. Choose the smallest legally deployable configuration that passes your own acceptance tests, and assess hosted APIs separately from downloadable weights.