Advertise here
#108

Inkling 2 Mini (local Q4)

Thinkingmachines Release: 2026-11-20 Tested on: 2026-11-22 16:20 thinkingmachines/inkling-2-mini-local::minimal
26.4B total (3.4B active)MoEOpen source
(low) (minimal)

Summary

Inkling 2 Mini (local Q4) scores 8.7 on AI BENCHY and ranks #108. It has 9.8 reliability, a 73.9% pass rate, ~$0.206 total cost, and 108.28s average response time.

What makes Inkling 2 Mini (local Q4) unique: It stands out most in Domain specific, where it ranks #1, while Combined is its weakest area at #16.

Model facts

Researched on 2026-11-22

Reported
Parameters
26.4B total (3.4B active)
Architecture
MoE
Availability
Open source
License
Apache-2.0

Same weights as Inkling 2 Mini, quantized by Unsloth to UD-Q4_K_XL GGUF and run locally through llama.cpp b12631. Multimodal looped sparse MoE: 8 of 128 experts per token, middle layers run three times (about 5.4B effective active parameters). Public Apache-2.0 weights with vLLM, SGLang and transformers instructions; use is also subject to the Thinking Machines Model Acceptable Use Policy. Training code and data are not published. Released 2026-11-20.

Score

8.7

Consistency

10.0

Total Cost (Current Price)

~$0.206

Total Output Tokens

219,084

Total Input Tokens

309,808

Estimated benchmark electricity cost

~$0.206

Tests Correct

Wrong Tests: 6

Attempt pass rate: 73.9%

Flaky tests

0

Flaky tests had mixed outcomes across runs (at least one pass and one fail).

Response Time (avg)

108.28s

Response Time (max): 1322.88s

Response Time (total): 2490.37s

Hamster playing table tennis

Prompt: Create a detailed SVG illustration of a hamster playing table tennis.

#108 Thinking Machines: Inkling 2 Mini (local Q4)

minimal
Cost
~$0.002
Time
67.8s
Tokens
2,348 tok

Charts

Choose the first model, then click a second model to open a side-by-side page.

Total Output Tokens

Score vs Total Output Tokens

Quick Compare

Category Breakdown

Category Score Consistency Tests Correct
Agentic 10.0 10.0
Anti-AI Tricks 10.0 10.0
Coding 10.0 10.0
Combined 6.4 10.0
Data parsing and extraction 10.0 10.0
Domain specific 7.7 10.0
General Intelligence 5.3 10.0
Instructions following 9.8 10.0
Puzzle Solving 6.0 10.0
Tool Calling 10.0 10.0
Trivia 3.0 10.0

Compared models