cd /news/artificial-intelligence/thinking-machines-releases-inkling-2… · home › topics › artificial-intelligence › article
[ARTICLE · art-174070] src=vibeleaderboard.ai ↗ pub= topic=artificial-intelligence verified=true sentiment=↑ positive

Thinking Machines releases Inkling 2 Mini, an open 26B laptop model

Thinking Machines released Inkling 2 Mini, an open 26B-parameter model under Apache 2.0 that runs on a MacBook, and Artificial Analysis scored it 20 on its Intelligence Index, level with 27B-class open models. Artificial Analysis measured a cost of $0.59 per task for Inkling 2 Mini versus $3.53, putting the mid-size open model at roughly a sixth of the cost per task.

key takeaway

Inkling 2 Mini is a 26B open model under Apache 2.0 that runs on a MacBook and scores 20 on Artificial Analysis's Intelligence Index, matching 27B-class open models at $0.59 per task versus $3.53.

read1 min views2 publishedNov 21, 2026

Artificial Analysis scores it 20 on its Intelligence Index, level with 27B-class open models, at $0.59 per task against $3.53. That puts a mid-size open model on a MacBook under Apache 2.0 for about a sixth of the cost per task. Read: Artificial Analysis scores it 20 on its Intelligence Index, level with 27B-class open models, at $0.59 per task against $3.53. That puts a mid-size open model on a MacBook under Apache 2.0 for about a sixth of the cost per task. Read: Terence Tao and Scott Aaronson both say verification is still pending. The claimed proof is a 17.4M-line Lean artifact built with Claude, so its acceptance rests on machine checking and on whether the formal statement matches the theorem. Read: It climbed from #30 to #1 since its September 22 launch. Claude Sonnet 5.5 and GPT-6.1 Sol each reached 22% of teams but handle under 2% of calls, so teams are trying new models widely while keeping most traffic on a cheaper default. Read: The attacks are executable and the code is public, so they can be checked rather than taken on trust. A model producing working breaks of submitted cipher designs moves AI-driven cryptanalysis from a claim to something security reviewers now have to plan for. Try: Before the fix, a client that controlled message history could replay a policy-denied tool call as approved and have it run. Apps using experimental_toolApprovalSecret relied on that signature as their human-in-the-loop guard. Read: Assistants such as Gemini Intelligence and Siri Onscreen Awareness read decrypted messages straight off the screen, outside the end-to-end encryption Signal provides. Signal now blocks them by default, with an opt-out.

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @thinking machines 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/thinking-machines-re…] indexed:0 read:1min 2026-11-21 · —