Tweet ghim
Four Qwen 3.5 models from @Alibaba_Qwen are now live on Tinker. Qwen 3.5 introduces hybrid linear attention that enables long context windows, as well as native vision input.
English
Tinker
69 posts
Tinker
@tinkerapi
I tink, therefore I am. Post-training API by @thinkymachines
San Francisco Tham gia Ocak 2026
1 Đang theo dõi3.2K Người theo dõi
tasty AND trained on Tinker >>>
Leonard Tang@leonardtang_
Hello MJ1: The World's TASTIEST Judge Model Agent verification is the bottleneck to AI's progress. The field's ability to verify visual output lags far behind that of text, especially in matters of ~taste~. So we built the world's tastiest multimodal judge model, MJ1.
English
Tinker đã retweet
Reasoning models waste tokens on easy questions. Instead of external scaffolding, @pranavxviswa trained token budgeting into the model itself and found that it got more accurate and while learning to think efficiently on its own.
x.com/pranavxviswa/s…
Pranav Viswanath@pranavxviswa
@aviral_kumar2 Great work Aviral! I also investigated this space using RLVR for reasoning models to adaptively budget their compute by problem difficulty using Thinking Machines’ Tinker (pranavviswanath.github.io/rlvr-reasoning/), found similar results!
English
New in the cookbook: Harbor RL trains models on real software engineering tasks inside sandboxed containers. The agent gets a bash shell, an instruction, and a test suite. If the tests pass, it gets the reward.
github.com/thinking-machi…
English
Tinker đã retweet
I joined @thinkymachines to work with @johnschulman2, @miramurati, @soumithchintala, @dchaplot, @alexgartrell, and others on the future of collaborative AI!
And I can confirm that we do have 45 lb weights.
If you're interested in training weights and/or Tinkering with us, join us! #join-us" target="_blank" rel="nofollow noopener">thinkingmachines.ai/#join-us
English
English
English
Tinker đã retweet
🚨 New: Integrating Harbor (@harborframework) for end-to-end Computer-Use evaluation(for Windows and Linux) at scale with @thinkymachines' Tinker, OSWorld, @daytonaio, and bare-metal servers.
We just added support for Computer Use, @tinkerapi, and OSWorld to Harbor - a framework for evaluating agents and generating RL training data by running large-scale rollouts across parallel sandboxed environments and collecting trajectories for SFT and RL.
Repo and blogpost below 👇
English
Tinker đã retweet
@rudzinskimaciej @Alibaba_Qwen pricing is live on the console and the homepage! thank you for flagging 🙇
English
@tinkerapi @Alibaba_Qwen Can you update the price to include thes models?
English
Four Qwen 3.5 models from @Alibaba_Qwen are now live on Tinker. Qwen 3.5 introduces hybrid linear attention that enables long context windows, as well as native vision input.
English
検索エージェントの質は推論より計画で決まる、という示唆が実装者には大きい。蒸留→GRPOの二段学習は、実務の改善ループを短くできる。
Tinker@tinkerapi
Contextual AI used Tinker to post-train the planning behavior for a search agent. They land on a two-stage training recipe: On-Policy Distillation and GRPO with a CLP reward. Read more 👇
日本語
@ethanelasky @thinkymachines pricing is updated in the console! for prefill / sample / train:
English
@thinkymachines any update on pricing for qwen3.5? the billing page hasnt been updated yet
English