Log inSign up
user avatar
Tinker
Thinking Machines
@tinkerapi
I tink, therefore I am. Post-training API by @thinkymachines
San Francisco
thinkingmachines.ai/tinker
Joined January 2026
1
Following
13.3K
Followers
RepliesMedia
  • Pinned
    user avatar
    Tinker
    Thinking Machines
    @tinkerapi
    Apr 9
    We’ve redesigned our docs with easy access to SDK reference, tutorials, support, and our newly updated cookbook---v0.3.0! Whether you’re writing your first training loop in Tinker or debugging async RL, we want to make it easier to find what you need.
  • user avatar
    Tinker
    Thinking Machines
    @tinkerapi
    1h
    Reasoning over speech requires privacy, low latency, and native audio processing instead of relying on transcripts. This was a major goal in training Inkling and Inkling-Small, and we’re proud of the results.
    user avatar
    Scale Labs
    @ScaleAILabs
    Jul 30
    Congrats to @thinkymachines on the release of Inkling-small! A smaller variant of Inkling, now live on our AudioMultiChallenge and MCP Atlas leaderboards. Inkling-small is tied for🥇on AudioMultiChallenge, scoring about the same as the larger Inkling despite the size difference.
    Image
  • user avatar
    Tinker
    Thinking Machines
    @tinkerapi
    7h
    The ability to mix compute platforms is novel and really useful for interpretability research. We're fans of what @GoodfireAI is building for researchers and love to see Tinker integrated into Silico.
    user avatar
    Ryan Panwar
    @RyanPanwar
    Aug 4
    Silico integrates with some of the best open weight training infra, such as @thinkymachines's Tinker. It also can run on your own hardware. And it can combine the two! Here it ran RL on Tinker while running a probe-based reward model on our B200 cluster.
    Image
    Image
  • user avatar
    Tinker
    Thinking Machines
    @tinkerapi
    Jul 27
    Inkling doesn't waste tokens getting to the punchline, we call that cheap humor.
    user avatar
    Adam Grenier
    @AKGrenier
    Jul 24
    I RL-trained @thinkymachines Inkling model to tell dad jokes, all from my laptop. Inkling is a 975B open-weights MoE. I fine-tuned it with GRPO on @tinkerapi into PunTune-0.6, which answers any topic with a focused, original dad joke. The jokes are... fine. But the finding
    Image
  • user avatar
    Tinker
    Thinking Machines
    @tinkerapi
    Jul 20
    Parameter-efficient fine-tuning isn't just cheap, it's what makes formal guarantees of model learning possible. Compress an RLVR update into a small LoRA and you can set a floor on how it will generalize to unseen data. Sharp paper from @maxYuxuanZhu , @rohanalur, and @ddkang.
    user avatar
    Daniel Kang
    @ddkang
    Jul 20
    New research from Bridgewater AIA Labs, UIUC, and MIT: we prove what we believe to be the first non-vacuous generalization bounds for reasoning LLMs on real-world problems. RLVR powers frontier reasoning capabilities yet its generalization to unseen data has remained an open
    Image