Skip to content
VECTOR WIREAI INTELLIGENCE
UTC

Inkling 2 Mini vs Qwen3.8 27B

Thinking Machines · released

Wins 3 of 5 areas

Agents · Reasoning · Long documents

Alibaba · released

Wins 1 of 5 areas

Images and charts

Inkling 2 Mini is the stronger all-rounder.Qwen3.8 27B is better at images and charts.

Scores updated · 32 tests both models report · How we compare

Where each one wins

Tests won in each of the five areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.

Images and charts and long documents rest on a single test each.

What it costs

Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.

Inkling 2 Mini$0.15 to read · $0.50 to write$0.65Price from Artificial Analysis
Qwen3.8 27B$0.50 to read · $3.00 to write$3.50Price from Alibaba's own price page

Inkling 2 Mini costs 81% less for the same work.

The biggest differences

The tests each model wins by the widest margin, up to three each. Scores are out of 100.

Where Inkling 2 Mini pulls ahead

  • Research-level physics problemsCritPt+26points aheadInkling 2 Mini31.4Qwen3.8 27B5.4
  • Finds hard-to-locate facts by browsing the webBrowseComp+17.4points aheadInkling 2 Mini79Qwen3.8 27B61.6
  • Questions about very long textsLongBench v2+14.6points aheadInkling 2 Mini66.9Qwen3.8 27B52.3

Where Qwen3.8 27B pulls ahead

  • Command-line tasks in a real terminalTerminal-Bench 2.1+8.2points aheadQwen3.8 27B79.8Inkling 2 Mini71.6
  • Harder college exam questions with imagesMMMU-Pro+1.4points aheadQwen3.8 27B76.3Inkling 2 Mini74.9

Every test, side by side

All 32 tests both models report. The winning score is in its model's colour; marks a score checked independently.

CodingEven

Full coding ranking

AgentsInkling 2 Mini

Full agents ranking

ReasoningInkling 2 Mini
  • CritPtInkling 2 Mini by 2631.45.4+26
  • ARC-AGI-2Inkling 2 Mini by 6.949.342.4+6.9

Full reasoning ranking

Images and chartsQwen3.8 27B

Full images and charts ranking

Long documentsInkling 2 Mini

Full long documents ranking

Other results23 tests, not counted

Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.

  • Misbehavior (Internal) ↓Qwen3.8 27B by 47.99.857.7+47.9
  • Covert misbehavior (Internal) ↓Qwen3.8 27B by 24.98.333.2+24.9
  • CritPt v2 FullInkling 2 Mini by 20.831.410.6+20.8
  • GraphWalks BFS 256KInkling 2 Mini by 19.369.750.4+19.3
  • JobBenchInkling 2 Mini by 16.249.633.4+16.2
  • CritPt v2 HardInkling 2 Mini by 13174+13
  • GDPval-AA 2.1Qwen3.8 27B by 108 rating points13191427+108 rating
  • Chartography originalInkling 2 Mini by 9.922.112.2+9.9
  • MRCR v2 8-needle 256K–512KInkling 2 Mini by 8.446.438+8.4
  • AutomationBench-AA v1.0.6Qwen3.8 27B by 6.441.848.2+6.4
  • DeepSWE 1.1Inkling 2 Mini by 648.242.2+6
  • DeepSearchQA (F1)Inkling 2 Mini by 5.685.379.7+5.6
  • ZeroBench (main) originalInkling 2 Mini by 51712+5
  • SWEBench Pro v2 (Hard)Inkling 2 Mini by 4.177.973.8+4.1
  • ComplexConstraints v1.1Inkling 2 Mini by 3.333.430.1+3.3
  • ARC-AGI-1Qwen3.8 27B by 384.587.5+3
  • BVBQwen3.8 27B by 2.368.470.7+2.3
  • MMMU Pro Standard 10Qwen3.8 27B by 1.474.976.3+1.4
  • OmniDocBench v1.6Inkling 2 Mini by 1.392.891.5+1.3
  • StrongREJECT (none)Qwen3.8 27B by 1.297.899+1.2
  • GDP (Surge AI)tie16.816.4tie
  • HLE-Diamondtie15.215.3tie
  • Terminal-Bench 4.0tie5.65.6tie

Questions people ask

Which is better, Inkling 2 Mini or Qwen3.8 27B?

Inkling 2 Mini wins three of the five areas where both have results: agents, reasoning and long documents. Qwen3.8 27B wins images and charts. They are level on coding.

Which is better for coding?

Neither. They win 1 coding test each of the 3 both models report, and 1 is a tie.

Which is cheaper?

Inkling 2 Mini costs $0.15 per million input tokens and $0.50 per million output tokens; Qwen3.8 27B costs $0.50 and $3.00. That makes Inkling 2 Mini about 81% cheaper for the same work.

How do you compare the two?

We use the 32 benchmark tests both models have published scores on. The verdict counts the 9 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 23 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.