stevibe ✓

@stevibe · joined 15 Jul 2009

LLM. Local AI addict. Building @BenchLocalAI DGX Spark workbench https://t.co/WN4i7q2y5m Builds things nobody asked for. Benchmarks things for fun.

27 800Followers

1 291Following

4 042Posts total

476KViews on collected posts

Latest posts

stevibe ✓@stevibe6 Sep

I asked GPT-6-Astra to build a first-person motion simulator. > A 240m cliff dive with underwater sun rays. > A glowing centrifuge tunnel. > A high-speed canyon run with banking turns. https://t.co/jTIvDhQGsl 1.8K views · 19 likes · 1 reposts · 0 replies Open on X →

stevibe ✓@stevibe6 Sep

@stevibe that looks dope 83 views · 1 likes · 0 reposts · 0 replies Open on X →

stevibe ✓@stevibe6 Sep

@stevibe worst possible render to audit too. a blurry glass blob has no ground truth, so nobody can prove the model did or didn't make it. free win forever. 41 views · 0 likes · 0 reposts · 0 replies Open on X →

stevibe ✓@stevibe5 Sep

@stevibe This is dope 105 views · 0 likes · 0 reposts · 0 replies Open on X →

stevibe ✓@stevibe5 Sep

If liquid glass is rendered in 3D, by GPT-6 Astra. https://t.co/3cquKeqRjO 8.9K views · 88 likes · 1 reposts · 7 replies Open on X →

stevibe ✓@stevibe4 Sep

I gave 4 "Flash" models the same job: redact the PII in an HR letter using an agentic mask tool. One of them cost 630× more than another. And it wasn't 630× better. Results 🧵 🥇 GLM 5.3 Flash $0.005, 15 tool calls Clean sweep. Overshot a bit (also redacted employee ID + https:/ 8.3K views · 73 likes · 8 reposts · 10 replies Open on X →

stevibe ✓@stevibe4 Sep

ChatGPT has been using 😂 a lot lately. https://t.co/EbhSRcBQ7H 1.6K views · 4 likes · 0 reposts · 1 replies Open on X →

stevibe ✓@stevibe26 Mar

@stevibe that's awesome. The qwen3.5-27b should win the year model medal. It is compact, huge on reasoning, fantastic on func tools. We are testing it on agemem --> 32 tools comprensive of auto-introspection where the models must manage the STM/LTM memories autonomsly. The qwen3. 3.2K views · 29 likes · 2 reposts · 1 replies Open on X →

stevibe ✓@stevibe25 Mar

I'm open-sourcing the framework behind this: ToolCall-15. Why build it? Every tool-calling benchmark I found was either: - Locked behind a research paper with no runnable code - Hundreds of abstract cases that don't map to real failures - Impossible to show on screen without a 17.9K views · 316 likes · 24 reposts · 16 replies Open on X →

stevibe ✓@stevibe25 Mar

Which local models can actually handle tool calling? I built a framework to find out. 15 scenarios. 12 tools. Mocked responses. Temperature 0. No cherry-picking. Tested every Qwen3.5 size from 0.8B to 397B, and since some of you asked after the distillation tests: yes, I https 433.9K views · 2.1K likes · 256 reposts · 112 replies Open on X →

Against accounts of the same size

7 posts from the last 90 days, next to the 10K–100K follower range. shown widely, but few of those viewers react.

Median views1 640this account924median for 10K–100K

Reach, %5.90%this account3.62%median for 10K–100K

Engagement, %1.09%this account1.52%median for 10K–100K

Metric	This account	Median for 10K–100K	Ratio
Median views per post	1 640	924	1.77×
Reach (views ÷ followers)	5.90%	3.62%	1.63×
Engagement rate	1.09%	1.52%	0.72×

Others in this range → Compare with another account → How these benchmarks are built →

Growth & engagement

How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.

Views per post

433.9K25 Mar

17.9K

3.2K26 Mar

1.6K4 Sep

8.3K

8.9K5 Sep

105

416 Sep

1.8K

Last 10 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.58%25 Mar

2.01%

1.04%26 Mar

0.30%4 Sep

1.13%

1.11%5 Sep

0.00%

0.00%6 Sep

1.20%

1.09%

Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 1.52%.

What the audience does

Likes52.6%2 634 in total

Reposts5.8%292 in total

Replies2.9%147 in total

Quotes1.5%75 in total

Bookmarks37.2%1 864 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.

Latest posts

Against accounts of the same size

Growth & engagement

Views per post

Engagement rate per post

What the audience does

Similar accounts