Toolbench Local models
Ollama
CJM Tested
Runs open models on your own machine behind a local API that coding tools can call.
- CJM score
- 6.4/10
- Last tested
- 2026-10-15
- Info checked
- 2026-10-15
- Website
- Visit Ollama
Overview
What it is
Ollama downloads and serves open weight language models on your own hardware with an OpenAI-style local API. Tools like Aider, Cline and Continue can point at it instead of a paid cloud model.
Who it is for
Builders who want private, free inference for coding agents and are fine trading quality and speed for zero per-token cost.
Hands-on test 2026-10-15
What we ran
Version: Ollama 0.40.4, Linux amd64, CPU only, Qwen2.5-Coder 7B Instruct Q4_K_M. Downloaded the release tarball and ran ollama serve. ollama pull failed on our network because the model download redirect resolved to a blocked address, so we downloaded the GGUF from Hugging Face and loaded it with ollama create and a Modelfile.
Create player.gd for a Godot 4 2D platformer. It must extend CharacterBody2D and use typed GDScript 2.0. Requirements: walk left and right with the ui_left and ui_right actions at 200 pixels per second, jump with ui_accept only when on the floor with a jump velocity of -400, apply gravity read from the physics/2d/default_gravity project setting, and call move_and_slide() with no arguments. No Godot 3 APIs.Result
- Served the model to Aider for the whole Godot test with no errors.
- Speed on our 8 core CPU: 6.2 tokens per second generating, 118.5 tokens per second reading the prompt. The standard prompt took 99.5 s end to end.
- Quality of the code is down to the model, not Ollama. See the Aider review for what this model produced.
Score
6.4 out of 10
| Area | Score |
|---|---|
| Setup | 6/10 |
| Output | 6/10 |
| Speed | 4/10 |
| Value | 10/10 |
| Game-dev fit | 6/10 |
Speed will be far better on a machine with a GPU or Apple silicon. We tested on a CPU only cloud box. The overall score is the plain average of the five areas.
Engines
Engine compatibility
Only engines we ran the tool with are marked as tested. Everything else says not tested yet, even where the vendor claims support.
| Engine | Status |
|---|---|
| Godot | Tested here |
| Unreal | Not tested yet |
| Unity | Not tested yet |
| Roblox | Not tested yet |
| Minecraft | Not tested yet |
Pricing
What it costs
Free to run locally, unlimited. Cloud plans: Free with starter credits, Pro $20 a month ($200 a year), Max $100 a month, Team $500 a month. Checked on ollama.com/pricing, 15 October 2026.
Prices change often. Check the vendor page before you buy.
Pros and cons
Good
- Single binary, served a 4.7 GB model with no GPU
- Imported a GGUF file directly when the built-in registry download was blocked by our network
- Worked as the backend for Aider with no extra config
Watch out
- CPU only inference is slow for agent loops
- Its own model download failed on our box because of a redirect rule, so we fetched the weights by hand
Alternatives