llama.cpp default port change to 9931
Parent: Mac local LLMs: Runtime selection and frontends · Published reference · snapshot 2026-10-05
↓ Facts as markdownall context files
PR 26508 "server: add notice for upcoming default port change 8080 --> 9931" was opened by a collaborator (ngxson) on 3 Aug 2026, approved by ggerganov the same day and merged as commit 0b14b87. It adds a notice only. The first title said 6631 and was corrected to 9931 in the same session.
These notes link each claim to its source. Some citations identify archived research reports that are no longer hosted on this site. A published reference means the content is available; it does not certify independent review or accuracy.Read the editorial policy and follow the sources before relying on a claim.
Facts
- PR 26508 "server: add notice for upcoming default port change 8080 --> 9931" was opened by a collaborator (ngxson) on 3 Aug 2026, approved by ggerganov the same day and merged as commit 0b14b87. It adds a notice only. The first title said 6631 and was corrected to 9931 in the same session. [source]
- The PR text says "Part of #26116" (transitioning llama-server from an on-demand command to a daemon process) and "Port gonna change from 8080 to 9931 (leetspeak of GGML)". [source]
- Issue 26116 (opened 25 Jul 2026, ngxson) specifies: `llama serve -hf` starts a llama-server in router mode if none exists, reuses an existing one, and downloads missing models; its open question was "which port? port 9931 == ggml". [source]
- The maintainer's clarification on 23 Aug 2026: `llama-server` will work as-is; the change is to the `llama serve` command; the only change to llama-server will be that `--port` defaults to 9931; setting `LLAMA_ARG_PORT=8080` in the shell profile restores 8080 without an explicit `--port`. [source]
- Stated reason: 8080 is a generic port other apps may use; the change is for UX, not technical. [source]
- The server README on master (fetched 2026-10-04) still documents `--port PORT` as "default: 8080" (env LLAMA_ARG_PORT) and says the server listens on 127.0.0.1:8080 by default; all its curl examples use 8080. [source]
- 25 Jul 2026 issue 26116; 3 Aug 2026 notice PR 26508 merged; 11 Aug 2026 LlamaBarn 0.40.0 moved its own default to 9931 "following llama.cpp, which is preparing to switch its own default too". [source]
- By 22 Sep 2026 issue 26116 was labelled stale and unresolved. [source]
- Anyone reading the 0.40.0 LlamaBarn note or the server startup notice can wrongly assume llama-server already listens on 9931. The default in the fetched README is still 8080; probes, plists and health checks must pin `--port` explicitly. [source]
- Users who reached issue 26116 from the PR read "transitioning llama-server ... to a daemon process" as llama-server being replaced; the maintainer said that is not the case. [source]
- A commenter objected that 9931 is not IANA-registered (ports below 49152 should be registered for application-aware firewalls). The maintainer replied that many apps use unregistered ports and users can set `--port`. [source]
- Two ports now coexist on a Mac running both llama.app (9931) and a hand-started llama-server (8080); clients with a hardcoded base URL hit the wrong one (inferred). [source]
- The LlamaBarn release note says llama.cpp is "preparing to switch its own default". The maintainer's comment says the change applies to the `llama serve` command and to llama-server's `--port` default. The README, as of the fetch, shows neither switched. Treat the plan as announced, not shipped. [source]
- Commenters (woof-dog, mcneiljt) pushed back on any move toward a daemon-only llama-server; the maintainer's answer limits scope to `llama serve`. [source]
- Which release will flip llama-server's default; no PR changing the default value itself was found in the issue search (search for "default port 8080" in titles returned only PR 26508). [source]
- Whether `llama serve` shipped in a release with port 9931 was not checked against release notes. [source]
- Merged PR 26508 (3 Aug 2026, commit 0b14b87) added a notice about the 8080 to 9931 default-port change; it did not change the default itself. [source]
- 9931 is described by the author as the leetspeak of GGML. [source]
- PR 26508 was approved by ggerganov and had 13 thumbs-up reactions. [source]
- Issue 26116 proposes `llama serve -hf` start or reuse a router-mode llama-server and auto-download uncached models. [source]
- Maintainer ngxson stated on 23 Aug 2026 that the only llama-server change is the `--port` default becoming 9931, and that LLAMA_ARG_PORT=8080 overrides it. [source]
- The stated rationale is that 8080 is generic and collides with other apps. [source]
- Issue 26116 carried the stale label from 22 Sep 2026. [source]
- The master server README fetched on 2026-10-04 still lists the default port as 8080. [source]
- Any service plist, health probe or client config for llama-server should pin `--port` or LLAMA_ARG_PORT so a later default flip does not break it (inferred). [source]
Corrections and disagreements
- CONTRADICTS: llama-macos-menu-bar-app.md line 53 ("Whether llama.cpp's own default port actually switched to 9931 was not confirmed"): it is now confirmed that a switch is announced via merged PR 26508 but not yet applied to llama-server in the README. [source]
Children
- No children recorded.