Local endpoint test: Ollama and Qwen3 8B

Version

Ollama 0.34.4 / qwen3:8b Q4_K_M

Current task

Local endpoint test: Ollama and Qwen3 8B

Validation

Two endpoints on an existing Windows installation passed synthetic checks; the first empty compatible response is retained as a failure.

Prerequisites

Prepare an authorized isolated test environment, pin dependencies and model versions, and back up before changes.

Estimated time

Learner time varies by environment; no measured duration is claimed.

Validation

Two endpoints on an existing Windows installation passed synthetic checks; the first empty compatible response is retained as a failure.

Validation

Confirm the request targets localhost and compare UI and API responses.

Limits & boundaries

Wrong port, unloaded model, insufficient resources or incorrect model ID.

Alternatives

ollama

lm-studio

Cost

Not verified

Price not verified; check software, model, API, hardware and hosting costs separately.

Implementation steps

1. 01
   Pin versions and model digest. Native /api/generate used CPU, 2048 context and 24 output tokens, returning BLUEFISH_OK.
2. 02
   Compatible chat first returned empty content with length at max_tokens=24; with 256 and provider-specific reasoning_effort=none, 2+2 returned 4.
3. 03
   Compatible chat used default GPU offload: observed context 4096 and about 5.58 GB VRAM. Temporary server stopped; no minimum VRAM or Agent tool-use claim.

Limits & boundaries

• Pass applies only to these versions, model and synthetic inputs; no soak, cross-device, regional, image or tool-calling test.

Tested by BlueFish · JSON ↗

Original sources

1. Ollama · Generate API ↗

   docs.ollama.com · 2026-10-16 · Official documentation

2. Ollama · OpenAI compatibility ↗

   docs.ollama.com · 2026-10-16 · Official documentation