Inkling-2-Small costs 3.7-4.9x less than Claude Sonnet 5.5 on structured pipeline tasks
HowDevelop · x · 2026-11-20
A hands-on benchmark of Thinking Machines' Inkling-2-Small (hosted on Nebius) as a frontier-model replacement for focused pipeline work:
- 6 real capabilities tested (vision, document processing, 3D generation, scoped coding, web research, knowledge work): 5/6 passed fully, vision matched 5/6 fields; all six runs cost $0.036 total
- Pricing: $0.30/1M input, $1.00/1M output — cheap even with 67% of output burned on hidden reasoning tokens
- Head-to-head with Claude Sonnet 5.5 on objectively verifiable scoped tasks:
- Scoped Python bug fix (6/6 tests both): Inkling $0.54/1K calls at 2.04s vs Sonnet $1.99/1K at 2.91s — 3.7x cheaper
- 3-way contract audit (5/5 checks both): Inkling $2.37/1K calls
Takeaway: focused, structured, verifiable tasks don't need a frontier model — a small model cuts cost 3.7-4.9x.
More from coding & agent
- Onlook goes open-source: local-first desktop app puts real products on an infinite canvas — ycombinator · 2026-11-20
- Kylon launches AI finance agents that paid $233K in supplier bills over four months — ying11231 · 2026-11-20
- Full automated GTA 6 content pipeline: Apify + ElevenLabs + Claude + Codex — eptwts · 2026-11-20
- TestSprite 4.0 launches as an independent test layer for AI-written code — Tegadesigns · 2026-11-20
- Dev builds video highlight feature with Gemini Neon, says Claude Opus 5.5 still wins on coding speed — vista8 · 2026-11-20
- VP of engineering: coding models report like bad direct reports — raw dumps, no TLDR — robleclerc · 2026-11-20