Inkling-Small
#8 en Modèles de langage open sourceThinking Machines · small · 16× · vu le 05 août 2026
69
Momentum
Inkling-Small is an open, natively multimodal Mixture-of-Experts language model from Thinking Machines Lab with 276 billion total parameters, of which 12 billion are active per Token. It natively processes text, image, and audio and generates text as output, supports a context window of up to 1 million Token, and is released under the Apache-2.0 license with full weights. According to Thinking Machines, it achieves near the performance of its four times larger sister model Inkling at significantly lower costs and was trained using On-Policy-Distillation from Inkling as teacher, among other methods.
Historique du momentum
07.05.05.08.
Fonctionnalités
| Key Benchmark (%) | 80.2% on SWE-bench Verified; Artificial Analysis Intelligence Index score of 40 |
| Context Window (Tokens) | Up to 1,000,000 tokens (model weights); Tinker serverless variant offers 256,000 tokens |
| License | Apache 2.0 (open weights) |
| Multimodality | Natively multimodal: processes text, image, and audio input, generates text output |
| Platform | Available via Tinker (fine-tuning & playground), Hugging Face (weights), and API hosting via OpenRouter and others |
| Price per 1M Tokens | $0.30 input / $1.20 output per 1M tokens (Tinker API standard pricing) |
| Release Date | July 30, 2026 |