Skip to main content

New model: tngtech/DeepSeek-TNG-R1T2-Chimera

Feature RequestsNew Model

Hi there,

R1T2 Chimera [1] outperforms R1 and is faster. After less than a week on OpenRouter [2], it processed more than 1.5 Bn tokens in a day. And it’s great!

I'm enquiring because in your post from April 28, you mentioned that DeepSeek was retired because it's too slow and the demand has trailed off. It seems that R1T2 Chimera reduces the issues from thinking traces by making them shorter and more useful. And the demand is increasing, perhaps because it has smaller memory requirements than Kimi K2.

Many thanks.

[1] https://x.com/tngtech/status/1940531045432283412
[2] https://openrouter.ai/tngtech/deepseek-r1t2-chimera:free

Status: Backlog1 comment

Log in to comment and vote

Comments1

  • Crimson Lollipop

    •

    Aug 8, 2025

    •

    Merged request

    •

    1 vote

    Look for an alternative to Venice Reasoning (Qwen QwQ)

    From what I've seen, Qwen 3 (Venice Large) already covers the capabilities of Qwen QwQ (Venice Reasoning), so it would be redundant to have Qwen QwQ.
    We could give Deepseek R1 Chimera a chance.