🤖Thinking Machines Unveils Inkling: Open-Source AI Model With 975B Parameters
Open-source Inkling aims to disrupt with its unique cost-performance balance
TL;DR
Inkling is Thinking Machines' first major language model under Apache 2.0 license, boasting high performance in benchmarks like SWE-bench Verified (77.6%) and VoiceBench (91.4%). It's designed to balance cost against performance through a novel mechanism.
Thinking Machines just dropped Inkling, its first major language model under an Apache 2.0 open-source license. This isn't your run-of-the-mill AI release; it's got 975 billion total parameters and is natively multimodal, capable of reasoning across text, images, and audio. If you're working on projects where cost and performance are key considerations, Inkling could be a game-changer. It achieves impressive scores in benchmarks like SWE-bench Verified (77.6%) and VoiceBench (91.4%), but it's the model’s ability to adjust its reasoning budget that sets it apart. Developers can programmatically control how hard the AI should 'think' before generating an output, allowing for a more flexible cost/performance curve.
Key Points
Inkling boasts 975 billion total parameters and 41 billion active parameters, making it a heavyweight in the AI model arena.
Achieved 77.6% on SWE-bench Verified and 91.4% on VoiceBench, showcasing its strength in software engineering and voice tasks.
Inkling's multimodal capabilities are evident with strong performance on MMMU Pro vision benchmark and MMAU audio processing tests.
The model includes a 'controllable thinking effort' mechanism, allowing developers to fine-tune cost vs. performance trade-offs.
Preview of Inkling-Small offers a lighter alternative optimized for low-latency and cost-sensitive workloads.
Why It Matters
If you're working on projects requiring high-performance AI models but constrained by budget or latency concerns, Inkling's unique approach to balancing cost and performance could be a game-changer. Its multimodal capabilities also make it appealing for applications like voice assistants or visual recognition systems where multiple data types are involved.
Frequently Asked Questions
Why does this matter?
If you're working on projects requiring high-performance AI models but constrained by budget or latency concerns, Inkling's unique approach to balancing cost and performance could be a game-changer. Its multimodal capabilities also make it appealing for applications like voice assistants or visual recognition systems where multiple data types are involved.
What happened?
Inkling is Thinking Machines' first major language model under Apache 2.0 license, boasting high performance in benchmarks like SWE-bench Verified (77.6%) and VoiceBench (91.4%). It's designed to balance cost against performance through a novel mechanism.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 2,075 builders reading daily.