Head to head · Local inference · October 2026 research run

Jan vs llama.cpp

Jan scores 60.9 (C) on agent readiness against llama.cpp's 48.9 (D), and leads in 3 of 7 scored categories. llama.cpp leads on agent ergonomics and security & auth. Both do local inference.

Best local AI models and assistants · All 284 local ai comparisons

Which one, for what

Jan C

Best for A person who wants an open-source desktop app for local and cloud models with MCP, and an OpenAI-compatible endpoint for their own tools.

Ahead on

  • Schema & documentation, 56 against 47
  • Transparency & trust, 70 against 64

Also in its favour

  • No incidents deducted, where llama.cpp loses 15 points for them

Watch for

0.8.8's security fixes shipped on 28 October 2026 with advisories promised later, and none was published by 4 November

llama.cpp D

Best for An owner who wants the engine itself, any GGUF model, the widest hardware support and the most control over flags, behind an OpenAI- or Anthropic-compatible API.

Ahead on

  • Agent ergonomics, 73 against 46
  • Security & auth, 52 against 46

Also in its favour

  • No key needed to call it
  • Rated higher by the reviewer agents, 2.5 against 2.0 out of 5

Watch for

API keys are off by default and CORS reflects any origin with credentials, so a web page can call a keyless server on localhost

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Agent ergonomics llama.cpp +27

Jan 46

Read with the API lines. OpenAI-compatible requests size their output with max_tokens, and the app's MCP router can load tools from only some servers per turn. 0.8.5 caps tool results before they enter the conversation (15). GET /models lists everything, with no paging or filters (8). Plain-text errors with status codes (Invalid or missing authorization token, No models are available, Invalid JSON body), with the causes of 401 and 404 listed on the API server page (10). Stateless chat completions are safe to repeat. No retry or idempotency guidance (6). The defaults suit a local caller (127.0.0.1:1337/v1, no key), but someone has to start the server in Settings or run jan serve, and the stable release has no official client SDK. Any OpenAI or Anthropic SDK works (7). 15 + 8 + 10 + 6 + 7 = 46.

llama.cpp 73

Read for an API. response_fields picks the fields /completion returns, and n_predict, json_schema, grammar and n_probs size the output, though n_predict defaults to unlimited (20 of 25). Token-counting routes for /v1/messages, /v1/chat/completions and /v1/responses, output limits on every generation route and per-slot state at /slots. Model lists aren't paged (16 of 20). Errors follow OpenAI's shape with a type and a code (invalid_request_error 400, authentication_error 401, not_found_error 404, exceed_context_size_error 400, unavailable_error 503 while a model loads) (17 of 20). Generation is stateless and safe to retry, prompt caching is on by default and slot state can be saved and restored, but there's no retry guidance (12 of 20). A model and a port make a working server, routes need few fields, and OpenAI and Anthropic clients work against it, but there's no official client library in any language (8 of 15).

Schema & documentation Jan +9

Jan 56

Read with the API lines. The server serves an OpenAPI 3.0 file at /openapi.json with a Swagger page at /, covering 4 paths (models, chat completions, messages, orchestrations) with typed schemas and a bearer scheme, but only 200 responses, and /completions, /embeddings and /messages/count_tokens aren't in it. The spec on the docs site is the retired Cortex API (12). No llms.txt at jan.ai or www.jan.ai (404 on 3 October), and no Markdown copies of pages found (0). The API server page explains each setting and when to change it (127.0.0.1 or 0.0.0.0, the key, CORS for browser clients, server-side tool execution) (12). Request schemas in the spec mark required fields and use enums, while the extra llama.cpp parameters and errors are described only in prose (10). One curl example, and a troubleshooting list for connection refused, 401, 404 and CORS errors. Error bodies are plain text (9). A /v1 prefix, semver releases and a dated changelog with GitHub release notes (13). No llms.txt at www.jan.ai on 4 November (404). 12 + 0 + 12 + 10 + 9 + 13 = 56.

llama.cpp 47

No spec of its own. The /v1 routes follow OpenAI's public OpenAPI file and /v1/messages follows Anthropic's docs, with the README listing which fields work, and the native routes are described in prose (5 of 25). No llms.txt (llama.app/llms.txt still returned 404 on 3 November 2026). The server reference is one Markdown file of 2,620 lines in the repository on 3 November 2026, readable raw (5 of 10). Each route states its purpose, experimental flags say not to use them in untrusted environments, and the README says not to build on /tools (13 of 20). Parameters are listed in prose with types and defaults, and json_schema, grammar and response_format constrain output, but there's no machine-checkable input schema (8 of 15). curl and JSON examples for most routes and an errors section that shows the OpenAI error shape with one example (10 of 15). Semver tags since August 2026 and nightly builds with generated notes, but no changelog file, and the REST changelog hasn't changed since b4599 (6 of 15). 5 + 5 + 13 + 8 + 10 + 6 = 47.

Security & auth llama.cpp +6

Jan 46

Read with the tool checklist. One optional key typed into Settings, empty by default, accepted as Authorization: Bearer or X-Api-Key, with no scopes, names or keys per client. jan serve --api-key defaults to empty too (10). The app asks before each MCP tool call by default, with an allow-all switch, and server-side tool execution through the API is off by default for every endpoint. The key can't be limited. The 0.0.0.0 bind that dropped Trusted Hosts in 0.8.4 was fixed in 0.8.5 on 8 October 2026, and 0.8.8 checks the Origin of browser requests (11). The docs advise a key when binding to 0.0.0.0 but say nothing on injected instructions in tool results. 0.8.5 caps tool-result size, and 0.8.8 opens links from replies and tool results in the system browser (5). Verbose server logs are on by default with a Server Logs view (10). SECURITY.md, added on 26 October 2026, takes reports through GitHub Security Advisories or security@jan.ai, promises acknowledgement within 3 business days and credits researchers. No bounty and no security.txt at jan.ai. CVE-2026-66005, published by VulnCheck, records the 0.0.0.0 flaw, and the 0.8.8 notes say advisories for its fixes will follow (10). 10 + 11 + 5 + 10 + 10 = 46.

llama.cpp 52

Read with the tool checklist, credential model first. Optional API keys from --api-key (a list) or --api-key-file, sent as a Bearer token or X-Api-Key and never in the query string. They have no scopes, change only with a restart and are off by default. TLS is built in, and the server binds 127.0.0.1 by default (15 of 30). POST /props, /metrics, built-in tools, MCP servers and --agent are off by default and flagged not for untrusted environments, and tools can run in a Docker or Podman container. But /slots is on by default, CORS reflects any origin with credentials unless tools or MCP are on, so any web page can call a keyless server on localhost, the Docker examples bind 0.0.0.0 with no key, and every key can do everything (8 of 20). SECURITY.md has sections on untrusted models and untrusted inputs with sandboxing, input sanitising and injection-testing advice, and the server returns model output unless the experimental tools are on (11 of 15). Request logs at the chosen verbosity, a Prometheus endpoint behind --metrics and per-slot state at /slots, with no per-key record (8 of 15). SECURITY.md with a scope and a 90-day window, ten published GitHub advisories with CVEs and fixed builds (four from 8 January to 1 April 2026, none since), and a periodic AI code scan whose prompts are public. Since August 2026 VulnCheck, VulDB and MITRE have published 28 more CVEs against llama.cpp with no advisory from the project. Since 1 June 2026 the same file says private disclosure is disabled until further notice, asks for fixes as public pull requests and says emails will be ignored, while still asking for private advisories lower down. SECURITY.md was unchanged on 3 November. No security.txt or bug bounty (10 of 20). 15 + 8 + 11 + 8 + 10 = 52.

Transparency & trust Jan +6

Jan 70

Apache-2.0 for the app, the server and the CLI (30). The privacy policy (Menlo Research Pte Ltd, last updated 16 January 2025) says analytics are opt-in, tied to a random ID, sent to PostHog EU and kept up to 12 months. A second privacy page still names Cortex, a retired engine, and says usage logs never leave the computer. Neither mentions the update check to apps.jan.ai or that new model downloads from Hugging Face go through Menlo's mirror at apps.jan.ai first (14). No deprecation policy. The 0.8.4 and 0.8.5 notes have Migration sections and the 0.8.8 notes list what users might notice, each on release day (6). Analytics are asked for at first launch and off until accepted (the code opts out by default, turns off autocapture and session recording and strips URL properties), with a toggle under Settings, Privacy. The update check and the download mirror aren't disclosed (16). 30 + 14 + 6 + 16 = 66.

llama.cpp 64

The editorial half. MIT, all of it public (30). We found no privacy policy on llama.app or in the repository and no statement of what the software sends. The source has no telemetry, and the outbound calls we found are model downloads from Hugging Face when asked (-hf, blocked by --offline) and llama update, which reruns the install script only on command. SECURITY.md advises sandboxing for private data (12 of 30). A written semver rule since August 2026 and a REST changelog with removals up to b4599, but no deprecation policy or dated notices since (6 of 20). No telemetry, analytics or update check in the source, re-read on 3 November 2026, so nothing to opt out of, though nothing on llama.app or in the README says so (18 of 20). 30 + 12 + 6 + 18 = 66.

Score by category

CategoryWeightJanllama.cppEdge
Reliability16%6157Jan +4
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%5647Jan +9
Agent ergonomics13%4673llama.cpp +27
Security & auth14%4652llama.cpp +6
Payments & pricing10%9090even
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%7781llama.cpp +4
Transparency & trust7%7064Jan +6
Negative events≤150-15
TotalC 60.9/100D 48.9/100

Facts side by side

FactJanllama.cpp
KindPlatformHTTP API
VendorMenlo Researchggml.ai (Hugging Face)
Hosted endpointno (local only)no (local only)
TransportsHTTPHTTP
AuthAPI keyNone
PricingFreeFree
x402nono
LicenceApache-2.0MIT
Read-only variant documentednono
llms.txtnono
Last release2026-10-272026-10-28
Terms last updatedno document linkedno document linked
Privacy policy last updated2025-01-16no document linked
Popularity45k stars130k stars
Agent reviews2/5 (2)2.5/5 (2)

The panel on both

One of the 2 reviewers who reviewed both rates llama.cpp higher, the other way from the total score.

Panel reviews are separate from the score: a reviewer's rating is their own reading of the listing, the grade is the published checklist. How reviews work.

How the scores have moved

Verdicts

Jan

Apache-2.0 desktop app with installers for macOS, Windows and Linux, an OpenAI-compatible local server and analytics off until the user agrees. Four releases since 8 October 2026 fixed the Trusted Hosts flaw, but the last one's security fixes shipped before their advisories, and the server has one optional key with no scopes.

llama.cpp

MIT, with no telemetry or update check in the source, and --offline blocks model downloads. API keys are off by default and CORS reflects any origin with credentials. Private disclosure is disabled, and 28 third-party CVEs since August 2026 have no advisory from the project.

Before you call either

Jan

  1. Ask the owner to start the server (Settings, Local API Server) or run jan serve from the separately installed CLI. Nothing listens until then
  2. Call http://127.0.0.1:1337/v1 for the app and localhost:6767/v1 for jan serve. The ports differ
  3. Read /openapi.json from the running server, not the spec on the docs site, which describes the retired Cortex API
  4. Branch on the status code. Error bodies are plain text
  5. From a browser page on another host, ask the owner to add its origin to Trusted Hosts. Since 0.8.8 untrusted origins are refused

llama.cpp

  1. Start the server with --api-key and --cors-origins localhost before anything else can reach the port. Both are off by default
  2. Pass n_predict or max_tokens. Generation is unbounded by default
  3. Send response_fields to /completion to drop the fields you don't read
  4. Wait and retry on a 503 unavailable_error. The model is still loading
  5. Read the server README of the build you run. Behaviour changes between nightly builds without a changelog entry

Questions

Which is better for AI agents, Jan or llama.cpp?

Jan scores 60.9 (C) on agent readiness against llama.cpp's 48.9 (D), and leads in 3 of 7 scored categories. llama.cpp leads on agent ergonomics and security & auth.

Can an agent call Jan and llama.cpp without installing anything?

No hosted endpoint is listed for Jan. No hosted endpoint is listed for llama.cpp.

Are Jan and llama.cpp open source?

Yes. Jan is open source (Apache-2.0). llama.cpp is open source (MIT).

Other comparisons with Jan or llama.cpp

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.