←
Tinker207 posts
Tinker avatar

Tinker

@tinkerapi

I tink, therefore I am Post-training API by @thinkymachines

◷ Joined January 2026▧ 5 Photos and videos
Jason Wei avatar
⟳ Tinker retweeted
Jason Wei@jaswwei·14h

1/ We taught a team of Inkling 2 Small agents to work together, communicate, and delegate work, without predefined roles on hard chess puzzles. Using our simple multi-agent RL recipe, we are able to turn pass@k into pass@1. Work with @emilianogl and @GabeVisotsky 🧵

1:14
Tinker avatar
Tinker@tinkerapi·19h

SAIR is keeping math open and collaborative with Open Math OS, letting scientists work together with open models. We're proud to support their work with Inkling 2 Small!

Inference and model access for Inkling 2 Small are sponsored by @thinkymachines through @tinkerapi, with zero data retention. Thanks to the Thinking Machines team for making this possible. Use Open Math OS now: openmathmodel.org/

Logan avatar
⟳ Tinker retweeted
Logan@loganzev·Nov 16

Excited to release The Trust Table. Been spending the last couple weeks learning how to use the new Tinker API, play with open source models and develop games for models to play and learn from. This is a personal benchmark I created as a research and product experiment. The goal was to learn how to fine-tune a model and explore a question I’ve been thinking about: do models know when they don’t know the answer? This is my first project of this kind, and it’s not a perfect study. The results are specific to this game, but I learned a lot from building it. Thank you to @thinkymachines and @tinkerapi for making the tools available to build and learn with!

0:06
Varun Gangal avatar
⟳ Tinker retweeted
Varun Gangal@VarunGangal·Nov 16

Can @thinkymachines new Inkling-2-Small not just play a video game level, but play and refine it many times over and learn to speedrun it? @akkikiki , myself and other co-authors from @UW and @PatronusAI ran one of the setups from our SpeedRunBench preprint from Sep this year to observe exactly that (arxiv.org/abs/2610.08076). Thanks to the @thinkymachines and @tinkerapi team for credits that enabled this study! Our benchmark measures whether agents can discover and then refine offline a fast enough policy through a level as they play more and more attempts, using only a single score after each attempt to adjust their policy. By design, we don't refine policy amidst episodes, or have multimodal frame understanding ... so our benchmark is more a read on planning, exploration, autoresearch etc than online videogame playing per se.. We used the "offline scratch" setting on SuperTux (SuperMario style game, Tux in place of Mario) level 1-1. Results: 1. Inkling-2-Small (256K, thinking effort 0.5) is an adept speedrunner, learning to clear the level after 1,988 frames, second only to GLM 5.3 (1,961 frames) in the open source family models we benchmarked .. 2. This put Inkling-2 ahead of Gemini 3.7 Flash, GPT-5.6-sol, Claude Opus 5, and DeepSeek-V4-Flash. See the attached bar graph for the final frames 3. Inkling-2-Small dramatically improves over its predecessor Inkling-Small which never learnt to fully clear the level, let alone improving at speedrunning it. Note for scale that Human world record iterated over the years for speedrunning this game is 1,619 frames...

Post attachment
0:50

Inkling 2 Small’s weights are now available for download on HuggingFace under an Apache 2.0 license. Make Inkling your own! huggingface.co/thinkingmachi…

Bhavish Pahwa avatar
⟳ Tinker retweeted
Bhavish Pahwa@BpHigh_BpLow·Nov 15

Post-trained Inkling 2 small on Railsim. Inkling 2 Small started at 0.119 on the 45 hard evaluation tasks, below a simple rule-based dispatcher (0.180). RL taught it to stop running out of words: 0.186. Switching to the long context model increased the score to 0.259 before any training, and 0.316 after a second round of reinforcement learning. Blog: huggingface.co/spaces/bpHigh… Space: huggingface.co/spaces/bpHigh… @thinkymachines @tinkerapi

0:31

Inkling 2 Small’s weights are now available for download on HuggingFace under an Apache 2.0 license. Make Inkling your own! huggingface.co/thinkingmachi…

Migel Tissera avatar
⟳ Tinker retweeted
Migel Tissera@migtissera·Nov 13

Tess-4.5-594B-A17B was fine-tuned on Thinking Machines' Tinker platform. What an amazing model you guys have trained! Thank you for releasing this on Apache-2.0. This is a legit frontier model, and I very much appreciate your commitment to open source! Thank you! 🙏🏾🚀 @thinkymachines @tinkerapi

Post attachment
emeraldant avatar
⟳ Tinker retweeted
emeraldant@emeraldant·Nov 13

We trained an open model to do weight-space programming. @thinkymachines’ Inkling 2 Small, post-trained on @tinkerapi, repairs broken networks from their raw weights in ~40 tokens, no reasoning. In the tokens it needed untrained to fix one, it now answers 207 and repairs 176.

0:17
Luc avatar
⟳ Tinker retweeted
Luc@lucrbvi·Nov 12

I built Modeline 584B A17B on top of Inkling 2 Small from @thinkymachines (which is now open-weight!) I used @tinkerapi to RL it on procedurally generated new levels of Celeste Classic and it got a +65% better score on CelesteBench! (my own bench) The model's CoTs are fun too

0:47
mohith avatar
⟳ Tinker retweeted
mohith@MohithAchu·Nov 12

Inkling 2 Small can draw pixel art. It just overthinks first. Base: ~3,100 tokens per icon, and half the time it ran out before drawing. After one small fine-tune on @tinkerapi: 74 tokens, 99% finish, correct drawings 40% to 68%. @thinkymachines @cHHillee

Post attachment
Evan Goldschmidt avatar
⟳ Tinker retweeted
Evan Goldschmidt@evangoldschmidt·Nov 9

we are observing this too. feels like a lot of performance to be squeezed from these models, and relatively straightforward to do with tinker + your own b200 cluster

it took me a while but I think I am coming around the the believe that finetuning (changing model's weights) is perhaps faster, cheaper and easier then prompt optimization. so many people are willing to run gepa and other prompt optimization, but for a reason I can't quite guess, much fewer are willing to finetune. you can literally just say to Claude Opus 5-5: i want a model that does X, find a paper about it, and some dataset. Make a better version using tinker from thinking machine. Start from inkling small model. All you have to do is put credits on think machines service (if you have not given your credit card and a browser to you coding agent XD).

Post attachment
Maxime Rivest 🧙‍♂️🦙🐧 avatar
⟳ Tinker retweeted
Maxime Rivest 🧙‍♂️🦙🐧@MaximeRivest·Nov 8

it took me a while but I think I am coming around the the believe that finetuning (changing model's weights) is perhaps faster, cheaper and easier then prompt optimization. so many people are willing to run gepa and other prompt optimization, but for a reason I can't quite guess, much fewer are willing to finetune. you can literally just say to Claude Opus 5-5: i want a model that does X, find a paper about it, and some dataset. Make a better version using tinker from thinking machine. Start from inkling small model. All you have to do is put credits on think machines service (if you have not given your credit card and a browser to you coding agent XD).

Post attachment
Tinker avatar
Tinker@tinkerapi·Nov 7

It’s encouraging to see how much tinker accelerated @msuiche- he’d previously burned 4600$ in gpu rental without finishing a single training step! He did some cool inkling 2 small experiments too. That’s tinker’s goal - so cheap and easy that there’s no reason to do it yourself.

Post attachmentPost attachment

Extremely bullish on American AI. Here is my (cyber) review Inkling 2 Small (Thanks @cHHillee & @brucedang ). I love the model, but I think I love the platform even more. I barely played with it and already feel how much potential it can unlock.

Post attachment
Zellige AI avatar
⟳ Tinker retweeted
Zellige AI@ZelligeAI·Nov 6

Tessera-27B-B-CJK-RL: our compressed-CJK coding agent, now trained with RL. AIME + HMMT 2026 11 point increase over non RL, Per agent turn it still uses 0.67x the base model's thinking. Open weights: huggingface.co/ZelligeAI/Tes… And our followup paper: zellige.ai/research/tessera-… Big thanks to @tinkerapi for sponsoring our continued research and development, We have much more in the works and have plans for more models and research in the near future.

Tessera-27B-B-CJK is a Qwen3.8-27B coding agent that thinks in compressed CJK, And saves tokens while doing so, Open weights, Apache-2.0. This is our first full tessera release, More info in the thread, check out our paper & hf: zellige.ai/research/tessera-… huggingface.co/ZelligeAI/Tes…

Tinker avatar
Tinker@tinkerapi·Nov 6

Kimi K3.1 Pro from @Kimi_Moonshot is live on Tinker today with 256K context. K3.1 Pro is the strongest open-weights model on several benchmarks of agentic coding and knowledge work, and is the largest model on Tinker today.

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 Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. Open weights by Nov 4, 2026. 🔗 API: platform.kimi.ai 🔗 Tech blog: kimi.ai/blog/kimi-k3-1-pro

Post attachment
echen avatar
⟳ Tinker retweeted
echen@echen·Nov 5

@johnschulman2 @miramurati our research team loved tinker btw! multiple RL experiments + evals in parallel w/o needing to manage gpus, and one line to switch model sizes