GUIDE · UPDATED OCT 10, 2026
Open-source LLMs for business
Open-weight models like Qwen3.8, DeepSeek V4 and GLM-5.3 now rival closed APIs on most business work, and you can run them on your own hardware or private cloud. You stop paying a third-party token markup and your data never leaves your perimeter. In August 2026, Alibaba even opened the weights of its flagship, Qwen3.8-Max.
By Lucas Page, The ShanghAI Guy · Updated October 10, 2026
Free. The 5 open-weight releases worth switching to, every week.
RUNS ON A LAPTOP
27B
Qwen3.8-27B runs on consumer hardware with quantization and, per Alibaba, matches Qwen3.7-Plus, a model ten times its size. Apache 2.0 license.
Source: Alibaba Cloud, August 2026.
THE DEPLOYMENT CHOICE
Self-host, hosted API, or closed API?
CRITERIA
HOSTED OPEN-WEIGHT API
SELF-HOSTED / PRIVATE CLOUD
CLOSED API
Cost
Low per token, from $0.16 per 1M input
Hardware and ops only, $0 token fees
Highest: $2 / $10 for Claude Sonnet 5.5
Data privacy
Depends on the provider
Full control, zero egress
Data leaves your perimeter
Setup
Minutes
Days to weeks
Minutes
Lock-in
Low: swap weights or providers
None
High: proprietary API and pricing
Best for
Proving a use case fast
Regulated data, high volume
Your hardest reasoning tasks
OPEN WEIGHTS WORTH RUNNING NOW
The October 2026 shortlist
MODEL
SIZE
LICENSE
USE IT FOR
Qwen3.8-27B
27B dense, 262K context
Apache 2.0
Laptop or single GPU. Coding and office automation.
GLM-5.3 (Zhipu)
744B MoE, 40B active, 1M context
GLM license (MIT-like)
Coding agents and security work.
DeepSeek V4-Pro
MoE, 1M context
See model card
Reasoning and air-gapped deployment (UnionPay).
Qwen3.8-Max
2.4T MoE, 95B active, 1M context
Qwen license
Flagship coding and professional work. Needs a GPU cluster.
Kimi K3 (Moonshot)
2.8T MoE, 1M context
Kimi K3 license
Large repos and long agent loops.
From each lab's model card and launch notes, checked October 10, 2026. Always read the license file before building a hosted product.
EVERY WEEK
The ShanghAI Radar
Each week we pick the 5 open-weight releases that can replace a paid tool, with specs, license, hardware needs and where to run them.
Start with the latest issue, then subscribe to get the next one in your inbox.
![[W39] 5 new open-weight models to replace closed APIs this week](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=720,fit=scale-down,onerror=redirect/uploads/asset/file/8db156ce-1366-442a-b319-08d127f9a8fa/Copie_de_ShanghAI_Focus.png?t=1790093611)
Switch-to-open-weight playbooks
Start here: the essential reads
[W39] 5 new open-weight models to replace closed APIs this week
Why US model restraints just made Chinese open weights unstoppable
Related guides: Chinese AI models for business · AI inference cost · For CTOs
FAQ
What decision makers ask us
Why use open-source AI instead of ChatGPT or Claude APIs?
You stop paying third-party token markups, keep customer data private, and can swap models without rewriting your stack. On high-volume work, savings typically run 60% to 80%.
Can open-weight models run on a normal office computer?
Small and mid-size ones can. Qwen3.8-27B runs on consumer hardware with quantization. Frontier-size models like Qwen3.8-Max (2.4T parameters) or Kimi K3 (2.8T) need multi-GPU servers or a hosted endpoint.
Can I use open-source LLMs commercially?
Usually, but read the license. Qwen3.8-27B is Apache 2.0 and GLM-5.3-Flash is MIT. Flagships often use custom licenses: GLM-5.3 requires a security review for model-as-a-service operators above $10B in revenue, for example.
Is open weight the same as open source?
No. Open weight means you can download and run the trained model; training data and code usually stay private. For most businesses, open weights is what matters: control, privacy and price.
Should I self-host or use a hosted API?
Start on a hosted API to prove the use case in days. Move to self-hosting or a private cloud once volume or data sensitivity justifies the infrastructure.
Know which open model to run before your competitors do.
Free every week: the open-weight releases worth switching to, tested for business.
Need it running on your private cloud? LeanAI handles the deployment.
Book a LeanAI cost audit