LikeGeeks

[LikeGeeks @likegeeks]

@PeasantSmith I mean the one with refusals and the non refusal model

[LikeGeeks @likegeeks]

@baltaci212 @coldniko Have you checked GLM5.3 flash? it's slower with maya engine, but don't know if smarter or not

[LikeGeeks @likegeeks]

@coldniko Caught the GUI! Hope it will be able to have web search ability and do deep research and other agent capabilities.

[LikeGeeks @likegeeks]

@PeasantSmith It would be great if you can make tests between those models to see how they perform

[LikeGeeks @likegeeks]

@Brjen Can you make another tests for these models

[LikeGeeks @likegeeks]

Hope they make Flash model we can run on consumer hardware

[Qwen Developers @QwenDevs]

alright guys, let's do both!

[LikeGeeks @likegeeks]

@DeepSeekHarness When you will add a revert (restore) button so anyone can revert changes made?

[LikeGeeks @likegeeks]

@PeasantSmith @spnova05 9 tokens/s is considered unusable if we speak about using it in something useful.

[LikeGeeks @likegeeks]

@jun_song Not all people have this beefy hardware to run GLM5.3 Flash

[LikeGeeks @likegeeks]

@ItsmeAjayKV @Aslan_ArtX but this shows Qwen3.8 flash next is better than GLM5.3, is this true?

[LikeGeeks @likegeeks]

@QwenDevs Big MoE

[LikeGeeks @likegeeks]

@karanC_12 @designwkarthick Does X pays for just impressions only? i mean just poeple view your tweets? or it requires interactions or reading your content?

[LikeGeeks @likegeeks]

@k2sbhai Just question brother, does X pays for just impressions? without post interactions or reading?

[LikeGeeks @likegeeks]

@coldniko @HououinTyouma Are you planning to support other MoE models?

[LikeGeeks @likegeeks]

If @GeminiApp didn't improve itself, human will replace it.

[LikeGeeks @likegeeks]

@IbrahimSait_ I'm using Qwen3.8-Flash-Next-GSQ-RCO with Strata engine on 16GB VRAM and 64 GB DDR5 and I'm getting 70-100 tokens/s generation

[LikeGeeks @likegeeks]

@coldniko @OdinsReisen You deserve it!

[LikeGeeks @likegeeks]

@seanhighness What about working with big coding projects? literally, I have 1000s of files in any project I works with.
Do you find that agent that do the perfect context compaction for 100s of times after it's full? or what do you mean?

[LikeGeeks @likegeeks]

@coldniko Will it support both the nvidea card and the builtin intel GPU? or this won't be beneficial?