Inkling 2 Small vs GPT-6 Luna
Both models appear in the same evaluation table on the official model card, which is the only place they are measured under identical conditions and reported by the same party.
Across 47 benchmarks where both models report a score, Inkling 2 Small is higher on 41 and GPT-6 Luna is higher on 6 . Inkling 2 Small has 594B total parameters with 17B active, and is priced at $1 per million output tokens on Tinker. Its largest margin is on AA-Briefcase v1.1. Its largest deficit is on Code2Video-AA.
Head to head
- Position
- Inkling 2 Small: the subject of this site · GPT-6 Luna: closed weights
- Bechmarks scored
- 47 shared rows, out of 71 total
- Record
- 41 wins · 6 losses · 0 ties
Largest differences
| Benchmark | Inkling 2 Small | GPT-6 Luna | Gap |
|---|---|---|---|
| AA-Briefcase v1.1 · Knowledge Work | 1,527 | 1,337 | + 190 |
| DesignArena 3D Design · Visual Creation | 1,401 | 1,213 | + 188 |
| GDPVal-AA v2.1 · Knowledge Work | 1,630 | 1,444 | + 186 |
| DesignArena Game Dev · Visual Creation | 1,428 | 1,261 | + 167 |
| DesignArena Web Apps · Visual Creation | 1,296 | 1,161 | + 135 |
| Code Arena · Visual Creation | 1,653 | 1,579 | + 74 |
| DesignArena Website · Visual Creation | 1,340 | 1,268 | + 72 |
| AA-Omniscience v1.1 · Factuality | 73.3% | 20.7% | + 52.6 |
| AA-Omniscience v1.1 · Factuality | 16.1 | -8.3 | + 24.4 |
| JobBench · Knowledge Work | 70.5% | 49% | + 21.5 |
Every shared benchmark
| Benchmark | Inkling 2 Small | GPT-6 Luna | Leader |
|---|---|---|---|
| Third Party Index | |||
| AA Index (v5.0) | 30.6 | 27.9 | Inkling 2 Small |
| Vals Index (v2.2) | 56.1% | 52.3% | Inkling 2 Small |
| Coding | |||
| FrontierCode v2.0 | 25.4% | 25.2% | Inkling 2 Small |
| DeepSWE v1.1 | 72.1% | 66.6% | Inkling 2 Small |
| Terminal Bench Science v0.1 | 5.7% | 8.6% | GPT-6 Luna |
| Terminal Bench v4.0 | 13.1% | 12.6% | Inkling 2 Small |
| Vibe Code Bench v1.1 | 81.6% | 81.7% | GPT-6 Luna |
| SWE-Marathon v1.1 | 38.8% | 28.1% | Inkling 2 Small |
| SWEBench Multilingual | 81.2% | 78.2% | Inkling 2 Small |
| SWEBench Multimodal v2 | 43.1% | 37.5% | Inkling 2 Small |
| SWEBench Verified | 82.2% | 76.1% | Inkling 2 Small |
| Autoresearch | |||
| BioMysteryBench | 70% | 61.5% | Inkling 2 Small |
| Search and Tools | |||
| MCP Atlas · Full / Public | 85.4% | 77.8% | Inkling 2 Small |
| Knowledge Work | |||
| AA-Briefcase v1.1 | 1,527 | 1,337 | Inkling 2 Small |
| GDPVal-AA v2.1 | 1,630 | 1,444 | Inkling 2 Small |
| JobBench | 70.5% | 49% | Inkling 2 Small |
| Finance Work | |||
| Finance Agent v2.1 | 67.5% | 56.8% | Inkling 2 Small |
| Excel Modeling Benchmark v1.1 | 69.5% | 70% | GPT-6 Luna |
| Tax Agent Bench | 65.8% | 58.9% | Inkling 2 Small |
| DAYJOB (Finance) | 14.3% | 6.8% | Inkling 2 Small |
| Apex-Accounting v1.1 · pass@1 / mean score | 14.5% | 10.8% | Inkling 2 Small |
| Diligence Bench · Best harness | 58.1% | 42.1% | Inkling 2 Small |
| Legal Work | |||
| Harvey LAB v1.1 | 21.7% | 4.2% | Inkling 2 Small |
| Legal Research Bench | 46.6% | 30.3% | Inkling 2 Small |
| Visual Creation | |||
| DesignArena Web Apps | 1,296 | 1,161 | Inkling 2 Small |
| DesignArena Website | 1,340 | 1,268 | Inkling 2 Small |
| DesignArena Game Dev | 1,428 | 1,261 | Inkling 2 Small |
| DesignArena 3D Design | 1,401 | 1,213 | Inkling 2 Small |
| Code Arena | 1,653 | 1,579 | Inkling 2 Small |
| Code2Video-AA | 1,020 | 1,030 | GPT-6 Luna |
| Visual Perception | |||
| PerceptionBench | 64.7% | 56.2% | Inkling 2 Small |
| ScreenSpot Pro · original / with Python | 83.8% | 81.1% | Inkling 2 Small |
| Humanity’s Sixth Sense | 33.3% | 21% | Inkling 2 Small |
| Visual Reasoning | |||
| GDP.pdf | 25.2% | 22.8% | Inkling 2 Small |
| MMMU Pro (Standard 10) | 81.6% | 79.7% | Inkling 2 Small |
| Charxiv RQ · original / with Python | 85.6% | 81.8% | Inkling 2 Small |
| Chartography · original / with Python | 36.7% | 32% | Inkling 2 Small |
| SAGE v1.1 | 88.6% | 83.4% | Inkling 2 Small |
| STEM Reasoning | |||
| CritPt v2 · Full / Hard | 55.7% | 65.1% | GPT-6 Luna |
| HLE-Diamond | 28.7% | 22.2% | Inkling 2 Small |
| HLE (text only) | 44.1% | 38.5% | Inkling 2 Small |
| Long Context | |||
| LongBench v2 | 70.6% | 65% | Inkling 2 Small |
| Factuality | |||
| AA-Omniscience v1.1 · score | 16.1 | -8.3 | Inkling 2 Small |
| AA-Omniscience v1.1 · non-hallucination rate | 73.3% | 20.7% | Inkling 2 Small |
| Instruction Following | |||
| ComplexConstraints v1.0 | 44.5% | 27.3% | Inkling 2 Small |
| Safety | |||
| DistressBench | 82% | 66.7% | Inkling 2 Small |
| Cybersecurity | |||
| CyberBench v1.1 | 72.7% | 76.3% | GPT-6 Luna |
Reading the comparison honestly
These numbers come from one table published by Thinking Machines, using third-party evaluations where those exist. That makes them comparable to each other, but it does not make them neutral: the vendor chose which benchmarks to run and how to present them, and a benchmark that a model loses has a way of not appearing.
The models are also not the same shape of thing. GPT-6 Luna sits in the closed weights tier of the card's own grouping, while Inkling 2 Small is the subject of this site. A win against a smaller model is less meaningful than a win against a larger one, and the table does not adjust for that.