Benz ON ⚙️

Meca Engineer, AI/ML Passion , Eternal learner.

Keep update on prices,TokenIndex : AI LLMs Tech Latest updates 🧮

[Benz ON ⚙️ @Benz_onttv]

📰 Some News:

🥇DeepSeek Flash is leading #1 and #2 position most used API on openrouter.

Some free intresting API : GLM 5.3 and Flash.

🪧Keep updated on : https://t.co/6O9fpd7qVW

[Benz ON ⚙️ @Benz_onttv]

@coldniko Yes, under 30 tok/s, is bit slow for serious usage. Keep it up

[Benz ON ⚙️ @Benz_onttv]

Hey,
You can now run frontier level model on you gaming PC. Of course not the full one, but these models out perfoms largely most of local model, even with smaller quants (IQ2 etc).
Thanks to @coldniko for Strata and @mhelcig for GSQ-RCO quants, Bravo👏

[Michael Helcig @mhelcig]

Run a 763B-parameter model locally. DeepSeek-V4.1-Flash-1017 on a 12GB GPU.

Frontier quality with GSQ-RCO quants, directly supported in the Strata inference engine.

Further info below. Much more to come. Follow for updates.

[Benz ON ⚙️ @Benz_onttv]

@mhelcig @DAlistarh @adotdad @coldniko Wow, need to try it ASAP. Thanks great work 👏

[Benz ON ⚙️ @Benz_onttv]

@OrcaRouter @coldniko I know you don't sleep well, but 🫣

[Benz ON ⚙️ @Benz_onttv]

@coldniko @mhelcig Oh, Hello ! there 👏

[Benz ON ⚙️ @Benz_onttv]

🆕 🌘K3.1 is here. Thats crazy, we now an open source frontier level model. Wish I can run it localy, may be one day.

The benchmarks look solid🧮.

Bravo @Kimi_Moonshot 👏

[Kimi.ai @Kimi_Moonshot]

Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.

🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.

Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on https://t.co/85jA71X9gw, Kimi Work, Kimi Code, and the Kimi API.

Open weights by Nov 4, 2026.

🔗 API: https://t.co/XCrgjXAqMw
🔗 Tech blog: https://t.co/5iPP0MBej6

[Benz ON ⚙️ @Benz_onttv]

@joe_ptrkv_ch Check profile you will find few replacements for Qwen3.6:35b.

[Benz ON ⚙️ @Benz_onttv]

@Kimi_Moonshot Hello there 👋

[Benz ON ⚙️ @Benz_onttv]

@PaulGugAI @Franzferdinan57 @Chris_Wozniczek @Krafton_AI Very recent, its a Korean model.

[Benz ON ⚙️ @Benz_onttv]

@MiaAI_lab I think we have real candidate for qwen3.6:35b
1. Raon 1 Mini Preview from @Krafton_AI
2. MiniCPM-V-4.7:35B-A3B from @OpenBMB
3. LFM2.5 24B-A2B from @liquidai
4. Spark-X2.6-4B really suprised me best speed/quality and impressive performance from @SparkLLM

[Benz ON ⚙️ @Benz_onttv]

@yvvlim Really good job, impressive even with Thinking OFF 👏

[Dongmin Park @dongmin_park11]

Thanks to the real review👏 can Raon-1 mini be the heir to Qwen3.6-35b? lets see 👀

[Benz ON ⚙️ @Benz_onttv]

@Kangwook_Lee Ok, we may have finaly the real heir to Qwen3.6:35b

Impressive RAW performance. 100% at SWE bench samples. With descent speed at high context, Q6 qant.

[Benz ON ⚙️ @Benz_onttv]

@TeksEdge And it delivers. Best model tested on my bench (for this size category). got 100% at SWE Bench, even Qwen3.8-Flash-IQ3_S didn't hit that score.

[Benz ON ⚙️ @Benz_onttv]

@Kangwook_Lee Well done 👏, top perfromance MoE model, I tested in a while. We'll see with actual daily usage.

[Benz ON ⚙️ @Benz_onttv]

@Kangwook_Lee Now, thats an impressive model.

Rank #1 in SWE bench, it got 100% first model I tested to get that socre.

Rank #1 at high context 128k ctx.

Descent speed ~35 tok/s for a Q6_K quant, even at 128k.

All of that raw performance, no MTP, no Thinking.

Well done 👏

[Benz ON ⚙️ @Benz_onttv]

@Kangwook_Lee Gonna add to my local low compute stack, will update here.

[Benz ON ⚙️ @Benz_onttv]

@bub_dot_md @SparkLLM @huggingface @ModelScope2022 @ollama @vllm_project @sgl_project Done, very good model for the size, its really impressive raw performance. Now, have to use it and see.

[Benz ON ⚙️ @Benz_onttv]

@Tech2Wild You will be suprised, we might actually have new low inference SOTA models. Will update once my benchmark finishes.