Local AI models, measured on one card.
Every model runs the same test suite on an RTX PRO 6000 with 96 GB: speed, coding, agentic work, puzzles and German. All results are free to read, no sign-up.
| Qwen3.8 Flash-NextQwen (Alibaba) · 125B parameters, 6B active | UD-Q4_K_XL GGUF | 123.6 | 109.3 | 4/5 | 7.0/7 | 5/5 | 2/2 | 1 Oct 2026 |
|---|
No model matches your search.
Speed: output tokens per second at 1k context with one request. Context: the same with 32k tokens in the prompt. Click a column header to sort.
How we test
Same prompts, same rules, same machine. Every model gets exactly one attempt per task.
Speed
Time to first token, prefill and output in tokens per second at 1k, 8k and 32k context, with one and with four concurrent requests.
Coding
Five tasks with one attempt each: a game, physics, a 3D scene, an SVG graphic and a small web app. The result runs offline in the browser and is recorded. Pass means: it runs and does the job.
Agentic
A customer case with four simulated tools: find the customer, check the shipment, issue a credit, write an email. Seven checks, three runs.
Puzzles
Five reasoning tasks with a single correct answer, from counting letters to combinatorics. Only the final answer counts.
German
Two writing tasks: a customer email with fixed requirements and an explanation for kids in exactly five sentences. We check the requirements and the language.
Settings
We pick the best quantization that runs sensibly on this machine. Quantization, runtime, sampling and context are listed on every model page.
Who runs the tests
Techat Lab is a project by Luca Hansmann. Luca also runs Techat, a platform with the AI assistant Techie, and this page promotes it. The measurements are independent of that: same suite, same machine.
The test machine
- GPU
- NVIDIA RTX PRO 6000 Blackwell, 96 GB VRAM
- CPU
- Intel Core i9-14900K
- RAM
- 64 GB
- Runtime
- llama.cpp (llama-server)
AI notice AI
The test suite collects the measurements automatically. Coding verdicts, conclusions and short texts are written with the help of Techie, Techat's AI assistant, from measurements and recordings. Such texts carry the “AI” mark.
Use these models in Techat
Techie, the AI assistant in Techat, normally runs on Qwen3.8 Flash-Next from this lab. It handles your website, email, bookings and contacts right in the chat.