I dont like my tokens go waste, so I have been using my ollama tokens with GLM 5.2 and it is fabulous.
I thought these models would never catchup claude but to be honest these are really good, I can say 70 percent. There is still 30 percent probably claude is smarter but I wrote a firefox extension and a small <a href="https://ozkanpakdil.github.io/godot-ottoman-sultans/" target="_blank" rel="noopener nofollow">game with godot</a> all has been fun
<img src="https://media.daily.dev/image/upload/s--kraxAEik--/f_auto/v1786879813/ugc/content_1d13eeb1-a9f9-4112-b5f0-774043166f0c?_a=BAMAMicg0" alt="image.png">
I started with kimi k3 but it was too expensive then I was going through the models, I tried kimi coder 2.7 that was also good but when I start using glm 5.2 others just fade away. Only problem I had, sometimes Github copilot VScode sends image read to GLM 5.2 which cant read so it breaks that session but in next prompt I say do not try to read the image and it works again :)
Also I had claude code experience, that was fun, it is little bit harder to read code from CLI, after couple of days I moved to VScode with github copilot, I just needed to define it like below
<pre><code>{
		&quot;name&quot;: &quot;ollama-local&quot;,
		&quot;vendor&quot;: &quot;customendpoint&quot;,
		&quot;apiType&quot;: &quot;chat-completions&quot;,
		&quot;models&quot;: [
			{
				&quot;id&quot;: &quot;glm-5.2:cloud&quot;,
				&quot;name&quot;: &quot;glm-5.2:cloud&quot;,
				&quot;url&quot;: &quot;http://localhost:11434&quot;,
				&quot;toolCalling&quot;: true,
				&quot;vision&quot;: true,
				&quot;maxInputTokens&quot;: 128000,
				&quot;maxOutputTokens&quot;: 16000
			}
		]
	}
</code></pre>
and of course I needed to run <code>ollama pull glm-5.2:cloud</code> then I manage to use vscode github copilot with glm 5.2 and finish all my tokens yesterday night 🤓 ollama is 20$ and it took me very long time to finish weekly tokens, so I think it is cheaper then github copilot or claude but I am not sure about GPT Luna, 2 weeks ago they drop its price 80 percent and I used to finish my work tokens very quick but Luna is very frugal so I am happy with that too.
How is your experience so far with the new models ?

Smiley Face

Özkan Pakdil

A developer shares a personal experience running local/cloud LLMs through Ollama, comparing GLM 5.2, Kimi K3, Kimi Coder 2.7, Claude Code, and GPT Luna for coding tasks in VS Code with GitHub Copilot. They describe configuring a custom endpoint for GLM 5.2, building small side projects (a Firefox extension and a Godot game), and note quirks like GLM 5.2 failing on image inputs sent by Copilot. Cost comparisons are made between Ollama's $20 plan, GitHub Copilot, Claude, and GPT Luna, ending with an open question to readers about their own experiences with newer models.

I manage to finish my weekly tokens

how are the ratelimits on ollama cloud? are they any better than other sub/providers?

I’d turn off vision for GLM 5.2 so Copilot stops sending it images

I was really thinking to shift from claude now i surely will.