ToolingPublished All news from that day
llm-d reaches 1.0, adding multi-tenant quality of service and zero-downtime upgrades for serving open models.
The maintainers' post says IBM Research serves up to 3,000 concurrent coding agents on 544 NVIDIA H100 GPUs using it.
Check our sources · 5 facts from 1 sourceThe llm-d maintainers published llm-d v1.0 on 20 October 2026.
The release post says it brings multi-tenant quality of service, zero-downtime upgrades and agent-aware serving on a stable v1 API. It says the project has more than 500 contributors from over 25 organizations, and that it serves open models through both vLLM and SGLang.
Its site identifies llm-d as a CNCF Sandbox project under the Apache 2.0 licence. The post reports that IBM Research serves up to 3,000 concurrent coding agents on 544 NVIDIA H100 GPUs using llm-d; that figure is the post's own account of a user's deployment.
Your reaction
We count reactions per story and day, never who reacted. The counts help us choose what goes in the monthly issue. If you are signed in, your own page shows yours too.
We count reactions per story and day, never who reacted. They help choose what goes in the monthly issue.
Check our sources
Every sentence above is checked against this source.
1 llm-d v1.0: State-of-the-art inference, ready for Day 2
Open the source archived copy-
The llm-d blog post "llm-d v1.0: State-of-the-art inference, ready for Day 2" is dated 20 October 2026.
citeOctober 20, 2026
-
The llm-d site says: "llm-d 1.0 is here! State-of-the-art inference, ready for production operations: multi-tenant quality of service, zero-downtime upgrades, and agent-aware serving on a stable v1 API."
citellm-d 1.0 is here! State-of-the-art inference, ready for production operations
-
The post says llm-d is "built by more than 500 contributors from over 25 organizations" and serves open models "through both vLLM and SGLang".
citebuilt by more than 500 contributors from over 25 organizations ... serving open models through both vLLM and SGLang
-
The site footer says llm-d is a CNCF Sandbox project and carries an Apache 2.0 License notice.
citellm-d is a CNCF Sandbox project ... Apache 2.0 License
-
The post says IBM Research self-hosts frontier open models on llm-d for coding agents and chip-design agents, "serving up to 3,000 concurrent coding agents on 544 NVIDIA H100 GPUs". The figure is the llm-d maintainers' account, not IBM's.
citeserving up to 3,000 concurrent coding agents on 544 NVIDIA H100 GPUs
Every sentence above is checked against this source. Each fact has its own address you can cite.
Source 1 · llm-d maintainers · 20 Oct 2026
llm-d v1.0: State-of-the-art inference, ready for Day 2
The llm-d blog post "llm-d v1.0: State-of-the-art inference, ready for Day 2" is dated 20 October 2026.
October 20, 2026
The llm-d site says: "llm-d 1.0 is here! State-of-the-art inference, ready for production operations: multi-tenant quality of service, zero-downtime upgrades, and agent-aware serving on a stable v1 API."
llm-d 1.0 is here! State-of-the-art inference, ready for production operations
The post says llm-d is "built by more than 500 contributors from over 25 organizations" and serves open models "through both vLLM and SGLang".
built by more than 500 contributors from over 25 organizations ... serving open models through both vLLM and SGLang
Show all 5 factsShow fewer facts
The site footer says llm-d is a CNCF Sandbox project and carries an Apache 2.0 License notice.
llm-d is a CNCF Sandbox project ... Apache 2.0 License
The post says IBM Research self-hosts frontier open models on llm-d for coding agents and chip-design agents, "serving up to 3,000 concurrent coding agents on 544 NVIDIA H100 GPUs". The figure is the llm-d maintainers' account, not IBM's.
serving up to 3,000 concurrent coding agents on 544 NVIDIA H100 GPUs
Topics
The morning email
On the mornings we publish: the three top stories and up to four short ones. Free.
We email you a link to confirm. An issue may include one sponsor, always labelled Sponsored · Advertisement. Our emails count opens and clicks, not who made them. Unsubscribe in one click. What we keep