BUSINESSDevelopments · · 2 min read · Source: Thinking Machines
Inkling 2 Small: open-weight model aims to match Gemini Flash
Thinking Machines, the AI startup founded by former OpenAI CTO Mira Murati, unveiled Inkling 2 Small on November 5, 2026. The company says the model weights will be released under the Apache 2.0 license after external safety testing, as early as next week.
My tool tip · AdGranola – AI meeting notesTry it free
Key facts
- Thinking Machines unveiled Inkling 2 Small on November 5, 2026: 594 billion parameters, 17 billion active.
- Weights are due under Apache 2.0 after external safety testing; at launch the model runs only on the Tinker platform.
- It scores 31 on the Artificial Analysis Intelligence Index – roughly level with Gemini 3.8 Flash, according to The Decoder.
What the model is
Inkling 2 Small is a mixture-of-experts model with 594 billion parameters, of which only 17 billion are active per request. It handles text and images, but not audio. At launch it runs only on Thinking Machines' own platform, Tinker, where you can try it and fine-tune it. Other providers are expected in the coming weeks.
Key numbers (company figures)
- Artificial Analysis Intelligence Index: 31 points, up from 13 for its predecessor. The Decoder puts that roughly level with Google's Gemini 3.8 Flash.
- Coding: 72.1% on DeepSWE, 86.3% on SWEBench Pro v2 (Hard).
- Cost: According to The Decoder, about $1.18 per task on the index, versus $7.77 for Gemini 3.8 Flash and $10.02 for Claude Sonnet 5.5.
The company also lists weaknesses openly: the model isn't always concise, doesn't always pick the best tool in large codebases, is less optimized for chat and trails closed frontier models on alignment evaluations.
Why open weights matter
If the weights ship as announced, companies can run and customize the model on their own servers or with a cloud provider of their choice. That appeals to organizations that can't or won't send their data to a US chat service – similar to Mistral Large 4, which arrived with open weights in late October.
What this means for you
- Open models are catching up: a freely usable model reaches the level of a current Google Flash model on the index.
- Cost per task drops sharply – important for companies using AI across many workflows.
- It's still an announcement: the weights are only free to download after safety testing.
How to use this
If your company is considering running AI itself, build a small test list of three to five real tasks and compare your current model with an open one. Decide based on those results, not benchmark tables.
Luigi's take
For many companies, the interesting question isn't which model ranks first, but whether they can run good AI while keeping control. Open models like Inkling 2 Small or Mistral Large 4 make that option more realistic. At the same time, these are company figures – benchmarks say little about how a model handles your texts, your workflows and your language.
My advice: don't rebuild anything yet. Take three to five real tasks from your day, such as answering a customer email or summarizing a proposal, and run them through your current model and an open one. If the open model keeps up on your tasks, it's worth talking to IT about running it yourself.
Source
Thinking Machines: Inkling 2 Small
More sources
- The Decoder – Einordnung mit Kostenvergleich nach Artificial Analysis
This article was researched and written with AI assistance. Responsible: Luigi Lugmayr.
DealPDFelement Black Friday: what 'up to 84% off' really saves
NewsInkling 2 Small: open-weight model aims to match Gemini Flash
NewsSoundCloud: AI music from ElevenMusic now publishes directly