Inkling 2 Small reference
API-only (weights pending) Apache 2.0 weights in 5 days (Nov 12, 2026) Model card checked Nov 6, 2026

Benchmark results

The complete evaluation table from the official model card, 71 rows across 13 models, plus the independent Artificial Analysis numbers. Scores are vendor-reported unless stated otherwise.

On the official model card, Inkling 2 Small holds the highest reported score in 16 of 71 benchmarks where it is scored, and the second-highest in 19. Its strongest categories are finance, legal work and visual design; it trails larger open-weights models such as Kimi K3 and DeepSeek-V4.1-Flash on software engineering benchmarks.

Independent measurement

Artificial Analysis
Intelligence Index
31 (v5.0.1) · rank 4 / 29
Time to first token
34.15s
Output speed
135 tokens/second
Time per task
18.5 minutes · median for comparable models 7 minutes
Cost per index task
$1.18
Output volume
154M output tokens across the index · 18.1B input tokens

Their summary: Inkling 2 Small is amongst the leading models in intelligence and reasonably priced when comparing to other models of similar price. It's also notably slow per task and very verbose.

Where it leads, where it follows

Highest reported score

  • BrowseComp 90.2%
  • JobBench 70.5%
  • EnterpriseOps-Gym 46.8%
  • EnterpriseBench-CoreCraft 71.8%
  • Finance Agent v2.1 67.5%
  • DAYJOB (Finance) 14.3%
  • DesignArena Web Apps 1,296
  • DesignArena Website 1,340
  • DesignArena Game Dev 1,428
  • DesignArena 3D Design 1,401
  • AA-3D 1,021
  • PerceptionBench 64.7%
  • ScreenSpot Pro 83.8%
  • BabyVision 68.6%
  • GDP.pdf 25.2%
  • AA-Omniscience v1.1 73.3%

Trails the leader

  • Sudo L7 14.1%
  • SWEBench Pro v2 (Hard) 86.3%
  • SWEBench Pro v2 (Full) 97.7%
  • PMPP-Hard 494.27
  • DeepSearchQA 90.9%
  • MCP Atlas 85.4%
  • Tau 3 Banking v1.0.1 40.4%
  • Diligence Bench 58.1%
  • Harvey LAB v1.1 21.7%
  • Humanity’s Sixth Sense 33.3%
  • MMMU Pro (Standard 10) 81.6%
  • Charxiv RQ 85.6%
  • MathVision 92.8%
  • Apex 2025 63.5%
  • HMMT Feb 2026 95.8%
  • LongBench v2 70.6%
  • DistressBench 82%
  • FORTRESS (adversarial) 83.8%
  • PatchEval 79.6%

Third Party Index

2 benchmarks
Third Party Index — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
AA Index (v5.0) 30.6 27.8 26.7 29.1 32.6 24.9 31.7 29.9 26.1 – 27.9 30.8 36.5
Vals Index (v2.2) 56.1% 50.9% 54.8% 53.8% 61.3% 48.6% 59.7% 57.8% 56.5% 53% 52.3% 56% 58.7%

Coding

14 benchmarks
Coding — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
FrontierCode v2.0 25.4% 14.2% 21.6% 20.8% 28.5% 7.6% 21.6% 24% 15% 17.3% 25.2% 18.7% 34.9%
DeepSWE v1.1 72.1% 63.4% 67.9% 73.7% 77% 60% 69% 71.9% 61.7% 68.5% 66.6% 73.8% 59.9%
Terminal Bench Science v0.1 5.7% 4.8% 4.3% 3.3% 12.9% 1.9% 9.5% 5.7% 4.3% 8.1% 8.6% 10% 20%
Terminal Bench v4.0 13.1% 32.8% 22.7% 25.8% 33.8% 8.1% 41.9% 34.9% 24.8% 12.6% 12.6% 19.7% 32.8%
Terminal Bench 2.1 · Inkling 2 Small: Pi harness 85.1% 84.3% 87.6% 81.7% 78.4% 76.4% 88.2% 89.9% – 88.3% – 89.4% –
Terminal Bench v2.1 (Terminus) 81.3% 84.3% – – – 76.4% 83.9% – – 85% – 87.6% –
Vibe Code Bench v1.1 81.6% 30.8% 79% 73.9% 87.7% 76.2% 78.1% 85.2% 81.6% 85% 81.7% 78.7% 90.4%
Sudo L7 14.1% – – – – – 15% – 13.9% 13.9% – 10.6% –
SWE-Marathon v1.1 38.8% 38.8% – – 40% – 42.5% – – 48.1% 28.1% – 50%
SWEBench Multilingual 81.2% 88.9% – 79.9% 82.9% 77.5% 87.6% – 79.7% 91.4% 78.2% 85.8% 83.7%
SWEBench Multimodal v2 43.1% – – – – – – 41.3% – 44.2% 37.5% 40.8% 44%
SWEBench Pro v2 (Hard) 86.3% – – 84.3% – 82.4% 84.3% – – 88.2% – 58.8% 52.9%
SWEBench Pro v2 (Full) 97.7% – – 96.6% – 95% 95.6% – – 97.7% – 94.9% 87.2%
SWEBench Verified 82.2% 92% – 81.1% 84.6% 80% 95.4% – 80% 93.4% 76.1% 80% 88.9%

Autoresearch

3 benchmarks
Autoresearch — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
PostTrainBench v1.2 31.9% 36.8% – – 30.5% – 34.5% – – 31.5% – – –
PMPP-Hard · AUC (task-hours, 0–10h) 494.27 – – – 497.36 – 430.96 – 478.39 453.58 – – –
BioMysteryBench 70% – 69.3% – – – – – – 72.6% 61.5% 62.2% 70.4%

Search and Tools

3 benchmarks
Search and Tools — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
BrowseComp · with context management 90.2% – – – – – – – 84.5% – – – –
DeepSearchQA 90.9% – – – – – – – – 95% – – –
MCP Atlas · Full / Public 85.4/85.0% 82% 62.9% 80.4% 83.2% 89.2/87.5% 84.2/83.6% 84.4% 84.5% 82.3/84.2% 77.8% 87.5% 81%

Knowledge Work

3 benchmarks
Knowledge Work — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
AA-Briefcase v1.1 1,527 1,451 1,487 1,476 1,570 1,279 1,503 1,512 1,348 1,497 1,337 1,203 1,574
GDPVal-AA v2.1 1,630 1,649 1,614 1,465 1,746 1,401 1,657 1,689 – 1,543 1,444 1,445 1,631
JobBench 70.5% 60.7% 61.2% 67.5% – 61.8% 61.4% 62% 55.2% 54.3% 49% – 39.2%

Enterprise Operations

3 benchmarks
Enterprise Operations — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
AutomationBench v1.0.6 45.6% 60.4% 52.3% – 66.6% 49.7% 62.2% 58.6% 57.8% 58.3% – 59.9% 35.4%
EnterpriseOps-Gym 46.8% 33.2% – – – – 36.4% – – 45.3% – – –
EnterpriseBench-CoreCraft 71.8% 46.7% – – – – 51.8% – – 43.1% – 58.5% –

Finance Work

7 benchmarks
Finance Work — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
Tau 3 Banking v1.0.1 40.4% 47.2% – – – – – – – 33.4% – – –
Finance Agent v2.1 67.5% 60.6% 63.3% 58.8% 62.9% 60.7% 65% 64.3% 61.7% 60% 56.8% 65.8% 65.4%
Excel Modeling Benchmark v1.1 69.5% 66.3% 68.9% 67.8% 69.2% 62% 69.1% 67.4% 68.7% 71.3% 70% 72.9% 65.5%
Tax Agent Bench 65.8% 62% 59.9% 62.7% 74.9% 52.1% 73.1% 64.9% 67.5% 68.7% 58.9% 66.8% 62.2%
DAYJOB (Finance) 14.3% 7.3% – – 12.5% – 11% – 4.8% 5.5% 6.8% 4.5% –
Apex-Accounting v1.1 · pass@1 / mean score 14.5/62.0% 12.3/53.7% – 13.5/60.0% 12.5/59.3% 13.8/58.0% 13.9/59.5% – 12.0/56.5% 9.5/56.7% 10.8/53.9% 15.5/58.8% 15.3/62.2%
Diligence Bench · Best harness 58.1% – – – 50.4% 59% 49.4% – – – 42.1% 51.1% 40.8%

Legal Work

2 benchmarks
Legal Work — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
Harvey LAB v1.1 21.7% 10.4% 13.3% 7.5% 14.2% 25% 14.6% 13.8% 15.8% 12.9% 4.2% 12.1% 4.2%
Legal Research Bench 46.6% 45.2% 38% 34.1% 45.2% 38.5% 49% 47.1% 35.6% 46.2% 30.3% 38.9% 43.8%

Visual Creation

7 benchmarks
Visual Creation — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
DesignArena Web Apps 1,296 – – – – – – 1,275 – 1,283 1,161 1,221 –
DesignArena Website 1,340 1,270 – 1,280 1,297 1,303 1,290 1,307 1,256 1,334 1,268 1,299 1,261
DesignArena Game Dev 1,428 1,276 – 1,312 1,261 1,270 1,303 1,315 1,225 1,351 1,261 1,305 1,319
DesignArena 3D Design 1,401 1,314 – 1,332 1,361 1,261 1,324 1,333 1,249 1,381 1,213 1,291 1,334
Code Arena 1,653 1,607 1,637 1,582 1,665 – 1,623 1,625 1,544 1,656 1,579 1,583 1,589
AA-3D 1,021 937 – – – – 782 977 – 982 – 965 994
Code2Video-AA 1,020 975 1,052 – – – 1,034 1,052 – 1,041 1,030 899 1,192

Visual Perception

4 benchmarks
Visual Perception — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
PerceptionBench 64.7% – – – 47.7% – – – – 58.5% 56.2% – –
ScreenSpot Pro · original / with Python 83.8/91.5% – – – – – – 56.6% 42.3% 73.8/83.2% 81.1/87.6% – –
Humanity’s Sixth Sense 33.3% 26.8% 25.4% – 30.1% 29.6% – 25.2% – 25.5% 21% 41.6% –
BabyVision · original / with Python 68.6/93.7% 53.4/82.6% – 91% 92.5% 44.3/80.2% – – 42.3/71.5% 53.4/85.7% – – –

Visual Reasoning

6 benchmarks
Visual Reasoning — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
GDP.pdf 25.2% 15.4% 9% 20.8% 17.2% 18.6% 11.2% 19.2% 17.6% 22% 22.8% 21% 20.8%
MMMU Pro (Standard 10) 81.6% 79% – 79.1% 80.6% 80.9% – 74.1% 78% 80.5% 79.7% 85.6% 79.1%
Charxiv RQ · original / with Python 85.6/87.7% 80.2/89.4% 66.2% 82.5% 77.0/85.6% 85.4/88.7% – 74.2% 83.2/87.9% 84.8/91.3% 81.8/85.3% 86.2/92.1% 84.4%
Chartography · original / with Python 36.7/80.4% 16.3/78.0% 7.5% 12.6% 17.9/83.0% 35.7/63.2% – 10.8% 21.2/56.6% 26.6/68.1% 32% 40.9/82.0% 46.4%
MathVision 92.8% 82.9% – – – 87.7% – – 88.9% 93.1% – – –
SAGE v1.1 88.6% 89.3% 83.3% 85.3% – 85.8% – 84.4% – 86.8% 83.4% 88.2% 89.8%

STEM Reasoning

6 benchmarks
STEM Reasoning — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
CritPt v2 · Full / Hard 55.7/35.0% 24.5% 29.5% 51.4/32.5% 50.6/25.0% 52.6/29.5% 57.4/33.5% 56.9/34.0% 50.3/27.0% 40.5% 65.1/42.5% 65.1/46.5% 64.0/43.5%
HLE-Diamond 28.7% 18.2% 13.7% 28.1% 25.3% 26.3% 17.9% 15.2% 17.3% 26.9% 22.2% 33.1% 29.6%
HLE (text only) 44.1% 39.9% 35.1% 48% 43.3% 43.9% 42.3% 49.4% – 46.9% 38.5% 47.8% 44.4%
GPQA Diamond 93% 91.2% – 86.7% 92.2% 89% 91.7% 89% 89.7% 93.5% – 95.3% 89.8%
Apex 2025 63.5% – – – – 32.3% – – – 65.6% – – –
HMMT Feb 2026 95.8% – – – – 93.9% – – – 97% – – –

Long Context

2 benchmarks
Long Context — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
AA-LCR v1.1 83.3% 80% – – – 78.3% 79.7% 86.3% – 88.7% – 81.3% 82.7%
LongBench v2 70.6% 68.2% 60.7% 60.9% 69% 68.7% 68.4% 68% 65.7% 69% 65% 73.7% 65.3%

Factuality

2 benchmarks
Factuality — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
AA-Omniscience v1.1 · score 16.1 0.7 -10.8 -6.4 0.9 25.8 6 -4.5 5.1 – -8.3 23.3 12.3
AA-Omniscience v1.1 · non-hallucination rate 73.3% – – 6.3% 35.9% 72% 61.1% 39.8% 71.8% – 20.7% 44.9% 65.8%

Instruction Following

1 benchmarks
Instruction Following — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
ComplexConstraints v1.0 44.5% 27.6% 22.1% 23.2% 50.4% 41.3% 30.5% 32.3% 36.5% 37.9% 27.3% 48.4% 36.8%

Safety

4 benchmarks
Safety — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
DistressBench 82% 72.7% – – – 87.7% 72.5% – 77% 73.5% 66.7% 70.7% 71.7%
FORTRESS (adversarial) 83.8% 73.3% – – 72.8% 92.4% 71.8% – 76% 72.8% – – –
FORTRESS (benign) 79.7% 96.4% – – 93.1% 84% 96.8% – 91.8% 96% – – –
StrongREJECT (none) 97.1% 97.8% 97.8% – 99.4% 97.1% 97.5% – 99% 99% – – –

Cybersecurity

2 benchmarks
Cybersecurity — as reported on the official model card, all 13 models. A filled dot marks the row's highest score, a hollow one the second-highest, exactly as the card annotates them. Missing entries are dashes.
Benchmark Inkling 2 Small GLM-5.3-Flash MiMo V2.6 Flash MiniMax-M3.1-Flash DeepSeek-V4.1-Flash-1017 Muse Spark Open GLM 5.3 MiMo V2.6 Pro Mistral Large 4 Kimi K3 GPT-6 Luna Gemini 3.8 Flash Claude Haiku 5.5
CyberBench v1.1 72.7% 67.9% 75.4% – – – 72.1% 72.9% – 75.2% 76.3% 43.8% 75.5%
PatchEval 79.6% 74.8% – – 80.4% 77.8% 69.6% – 76.1% – – – –

How to read this table

Every score here is what the vendor published. Thinking Machines report that evaluations ran at effort 0.99 (xhigh) and temperature 1.0, with a 1M-token maximum trajectory limit, and that where a third party publishes an evaluation, the third party's number is used rather than an internal one. That source list includes Artificial Analysis, Scale AI, Vals, Datacurve, DesignArena, Surge, Mercor, Thoughtful and Arena.

The rank dots are the card's own annotations, not ours. They compare across the full row of 13 models, which means a benchmark can legitimately have no marked winner if the leading score came from a model that was excluded from that comparison. Unit notes: raw score, %, task-hours (AUC).

Comparative pages: vs Kimi K3, vs DeepSeek-V4.1-Flash, vs GLM 5.3, or every model on the card. The first-generation Inkling models are not in this table; their size and price are compared on the pricing page.