4 Comments
User's avatar
Paul Graham's avatar

It's not free, and I've already been charged money when using it!

Expand full comment
GW's avatar

Spectre was a total disappointment for us. We have a series of prompts that we use for testing each LLM and this model failed miserably. Horrible at tool calling. Hallucinating more than any model we have every worked with. Very hard to steer. Something is wrong with this model, it needs to be fixed. I don't understand the hype, because it fails at any real world challenges.

Expand full comment
Joe C's avatar

It was a flop for me. My first attempt with a prompt of 2 lines, burnt 17 million tokens and didn't produce a working result.

I tried with a very basic HTML website which it managed, but looked bad. And the 3rd attempt at a simple web tool to edit image sizes failed miserably.

Expand full comment
Akshay Joshi's avatar

Spectre was disappointing for me as well. Its output was far below what I would expect from a model positioned to follow Claude Opus 4.5, the current benchmark for agentic AI code generation. I even stayed up until 2 a.m. trying to get it to work, wondering if I had done something wrong, especially since Kilo Code claimed it came from one of the top ten AI labs. Ultimately, it did not meet expectations.

Expand full comment