I’ve been trying to develop a CAT tool for personal use for almost a week now. I wrote an incremental integration test with 48 tasks to track progress and follow a TDD-style development workflow. [2025-07-09_21-03-26] I’ve hit a wall at Task 20 — none of the available LLMs have managed to pass it…

Task: Passing a Pytest in a Test-Driven Development workflow The test cases are offered by Claude 4 when working as a QA Agent in the another repository. Gemini 2.5 Pro Gemini 2.5 Pro failed to handle it for $0.75; launched in an already ongoing long chat Grok 4 Gemini only wrote tests test…

[image] Artemonim: Right now, Grok 4 is kind of broken. It almost always stops either right after the prompt or mid-execution. He finishes the job in a one-shot style, but maybe I just got lucky. Still, I want a proper chat experience. I’m having this experience as well. With Claude 4 if I as…

[image] roy7: With Claude 4 if I ask it to run the tests and try to fix the one that’s broken, it’ll re-run the test each time it tries something new and debug it. Strange. I only have such problems with Grok 4 and Gemini 2.5 Pro.

Yeah. To be clear, I’m saying Claude does this correctly and Grok isn’t.

While refactoring the current Agent Enforcer , I gave Grok 4 the rundown on the MCP methodology—but totally forgot to mention FastMCP. So what did Grok do? That’s right—it just dove into the venv and read the FastMCP code on its own :smiling_face_with_sunglasses: [2025-07-16_09-17-05] Just look at …

Let me explain what happens for those learning: experienced users try to include all the required context at start and the agent gets missing context as a last resort, Grok as a reasoning agent(right now the best) did that but included unneeded context which is called context poisoning and output er…

I stopped supplying a lot of context after Sonnet 4 came out. Now my rules are like 10 lines with core principles. Prompts are usually one sentence.

Gemini always has a long thinking phase at the first prompt in the dialogue. I assume that he plans better if the necessary context is passed to him with the first prompt, rather than through subsequent read_tool. Grok 4 can generally act without context.

Yes it does depend on the model as well.

Keep in mind there’s always context when using Cursor, that’s why @condor can one-shot without giving context, Cursor adds file structure(if ticked in settings), git graph(if ticked), context related to the prompt through Vector search on your codebase index, your mcp tool descriptions and system pr…

(Continuously Updated) My Real-Time Review of Grok 4

Discussions

normalnormie (Norman) July 16, 2025, 10:12am 47

Well, the absolute leader in reasoning, to code we need agentic coding and it’s the worst between SOTA models, it needs a lot of work for tool use, but they’re pretty active on it:

Topic		Replies	Views
Grok free on Cursor - Feedback needed Discussions	71	5623	September 24, 2025
`sonic` Ghost Model Discussion Release Discussions	75	7050	August 30, 2025
Gemini 3.0 Pro - Out Now! Release Discussions	113	15784	December 6, 2025
How To Optimize Your Usage: The Best AI Models to Use, version 2.2 Guides	28	4135	December 9, 2025
Is it just me, or is GPT-5's logic for code incredible? Discussions	27	9494	September 28, 2025

(Continuously Updated) My Real-Time Review of Grok 4

Related topics