# Thinking Machines Inkling 2 Small

As of November 14, 2026, Thinking Machines Inkling 2 Small ranks #63 of 118 models overall with 56.9 ±9.3 on the LLMBenchmark Index. Inkling 2 Small ranks #49 of 122 in reasoning, its strongest category. Thinking Machines Inkling 2 Small costs $0.30 per million input tokens and $1.00 per million output tokens. Inkling 2 Small was released on November 9, 2026.

Page: https://llmbenchmark.ai/models/thinkingmachines-inkling-2-small
Data from November 14, 2026. Every figure below names the board it was read from and the day.

- Maker: Thinking Machines
- Released: November 9, 2026
- Price per million input tokens: $0.30
- Price per million output tokens: $1.00
- Context window: 1.05M tokens
- Overall rank: 63 of 118, by the LLMBenchmark Index
- LLMBenchmark Index: 56.9 ±9.3 (90% margin, from 4 of 7 inputs)
- Artificial Analysis Intelligence Index: 30.6
- Vibe Coding Index: 48.9 (from 3 of 6 coding benchmarks)

Benchmark results (51 of 222):

- LMArena WebDev: 1647 Elo (source: LMArena WebDev, read November 14, 2026)
- Vals Index: 56.1% (source: Vals Index, read November 14, 2026)
- AA Intelligence Index: 30.6 (source: Artificial Analysis (via OpenRouter), read November 14, 2026)
- Terminal-Bench 4 (Vals AI): 12.6% (Mini-SWE-agent) (source: Vals AI boards, read November 14, 2026)
- Vibe Code Bench (Vals AI): 81.5% (OpenHands) (source: Vals AI boards, read November 14, 2026)
- ProgramBench (Vals AI): 0.5% (source: Vals AI boards, read November 14, 2026)
- SREBench (Vals AI): 1.9% (source: Vals AI boards, read November 14, 2026)
- SWE-bench Multimodal (Vals AI): 43.1% (source: Vals AI boards, read November 14, 2026)
- CyberBench (Vals AI): 72.7% (source: Vals AI boards, read November 14, 2026)
- Tax Agent Bench (Vals AI): 38.9% (source: Vals AI boards, read November 14, 2026)
- Legal Research Bench (Vals AI): 46.6% (source: Vals AI boards, read November 14, 2026)
- Harvey's Legal Agent Benchmark (Vals AI): 21.7% (source: Vals AI boards, read November 14, 2026)
- BioMysteryBench (Vals AI): 70.0% (source: Vals AI boards, read November 14, 2026)
- Terminal-Bench-Science (Vals AI): 2.9% (Mini-SWE-agent) (source: Vals AI boards, read November 14, 2026)
- Vals Science Index: 32.7% (source: Vals AI boards, read November 14, 2026)
- MysteryMechanism (Vals AI): 25.2% (source: Vals AI boards, read November 14, 2026)
- SAGE (Vals AI): 88.6% (source: Vals AI boards, read November 14, 2026)
- Humanity's Last Exam (Artificial Analysis run): 44.1% (source: Artificial Analysis, read November 14, 2026)
- SciCode: 52.3% (source: Artificial Analysis, read November 14, 2026)
- Long context reasoning (AA-LCR): 83.3% (source: Artificial Analysis, read November 14, 2026)
- Terminal-Bench 4.0 (Artificial Analysis run): 13.1% (source: Artificial Analysis, read November 14, 2026)
- FrontierCode 2.0: 25.4% (source: Artificial Analysis, read November 14, 2026)
- HLE-Diamond (Artificial Analysis run): 28.7% (source: Artificial Analysis, read November 14, 2026)
- GDPval-AA: 1630 Elo (source: Artificial Analysis, read November 14, 2026)
- AA-Omniscience: 16.1 (source: Artificial Analysis, read November 14, 2026)
- CritPt v2, hard subset (Artificial Analysis run): 35.0% (source: Artificial Analysis, read November 14, 2026)
- APEX-Agents: 46.8% (source: Epoch AI benchmarking hub, read November 14, 2026)
- ARC-AGI: 96.5% (xHigh) (source: Epoch AI benchmarking hub, read November 14, 2026)
- ARC-AGI-2: 81.7% (xHigh) (source: Epoch AI benchmarking hub, read November 14, 2026)
- LMArena Text: 1403 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: coding: 1455 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: hard prompts: 1421 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: instruction following: 1408 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: creative writing: 1348 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: English: 1418 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: non-English: 1390 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: Russian: 1377 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: multi-turn: 1408 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: longer queries: 1403 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: expert: 1478 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: hard prompts in English: 1424 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: software and IT: 1443 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: science: 1409 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: business and finance: 1415 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: writing and language: 1381 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena Text: entertainment and media: 1359 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena WebDev: React: 1660 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- LMArena WebDev: HTML: 1661 Elo (source: LMArena (Hugging Face dataset), read November 14, 2026)
- MathArena: 42.4% (source: MathArena, read November 14, 2026)
- MCP Atlas: 85.4% (source: Scale SEAL leaderboards, read November 14, 2026)
- SWE-Bench Pro V2: 97.7% (source: Scale SEAL leaderboards, read November 14, 2026)