Американский AI-стартап Thinking Machines Lab открыл веса языковой модели Inkling 2 Mini, которая по бенчмаркам достигает уровня более крупной Qwen3.8-27B. Модель вышла 20 ноября 2026 года под лицензией Apache 2.0, веса доступны на Hugging Face.
Контекстное окно составляет до 1 млн токенов. MoE-архитектура на 26,4 млрд параметров, из которых активны 5,4 млрд. В каждом слое токен маршрутизируется на 8 из 128 экспертов.
Поддерживаются Hugging Face Transformers, vLLM, SGLang, TokenSpeed, NVIDIA Dynamo, Unsloth (включая GGUF), llama.cpp, Ollama и Exo Labs. Также веса доступны через платформу дообучения Tinker, где есть API и playground.
Бенчмарки
| Бенчмарк | Inkling 2 Mini | Gemma 4 31B | Muse Glimmer | Qwen3.8-27B | Nemotron 3.5 Super VL |
|---|---|---|---|---|---|
| Параметры, млрд (активные / всего) | 5/26 | 31/31 | 30/30 | 27/27 | 12/125 |
| Сторонние индексы | |||||
| AA Index (v5.0) | 20.1 | 4.4 | 6.3 | 20.1 | 13.6 |
| Vals Index (v2.2) | 46.5% | – | – | – | 29.8% |
| Кодинг и автоисследования | |||||
| FrontierCode v2.0 | 15.1% | 0.0% | 1.5% | 14.9% | 6.1% |
| DeepSWE v1.1 | 48.2% | 0.3% | 4.4% | 42.2% | 24.7% |
| Terminal Bench v4.0 | 5.6% | 0.0% | 0.5% | 5.6% | 0.0% |
| Terminal Bench v2.1 | 71.6% | 43.4% | 51.7% | 73.0% | 60.7% |
| SWEBench Pro v2 (Hard) | 77.9% | 14.2% | 44.3% | 73.8% | 44.1% |
| SWEBench Multilingual | 73.7% | 43.0% | 66.9% | 70.0% | 67.7% |
| SWEBench Verified | 79.2% | 46.1% | 76.0% | 77.8% | 71.1% |
| Vibe Code Bench v1.1 | 77.9% | – | – | 64.9% | 3.5% |
| PMPP-Hard (AUC, 0–10h) | 455.82 | – | – | 262.50 | – |
| Агентный поиск, tool use и knowledge work | |||||
| BrowseComp | 79.0% | 37.0% | 63.9% | 61.6% | 78.9% |
| DeepSearchQA | 85.3% | 61.7% | 74.6% | 79.7% | 77.7% |
| MCP Atlas (Public) | 82.7% | 54.2% | 75.5% | 77.3% | 67.7% |
| AA-Briefcase v1.1 | 1263 | 369 | 478 | 1391 | 999 |
| GDPVal-AA v2.1 | 1319 | 627 | 792 | 1427 | 1162 |
| JobBench | 49.6% | 8.2% | 16.1% | 33.4% | 36.5% |
| AutomationBench-AA v1.0.6 | 41.8% | 4.9% | 6.8% | 48.2% | 14.8% |
| EnterpriseOps-Gym | 42.0% | 28.4% | 34.7% | 44.2% | 28.1% |
| Vals Finance Index | 50.1% | – | – | – | 41.0% |
| AA-AnalystAgent | 46.3% | – | 20.0% | – | – |
| Diligence Bench | 48.6% | – | 40.9% | 46.4% | 45.0% |
| Зрение | |||||
| GDP.pdf | 16.8% | 6.0% | 11.8% | 16.4% | 8.0% |
| PerceptionBench | 57.5% | 40.7% | 42.8% | 52.4% | 36.8% |
| MMMU Pro (Standard 10) | 74.9% | 73.4% | 74.3% | 76.3% | 72.3% |
| OmniDocBench v1.6 | 92.8% | 82.5% | 90.4% | 91.5% | 77.9% |
| ScreenSpot Pro (оригинал / с Python) | 76.5 / 87.7% | 5.9% / – | 75.4% / – | 57.6 / 78.9% | 43.5% / – |
| Chartography (оригинал / с Python) | 22.1 / 72.5% | 12.0 / 11.8% | 17.4 / 25.1% | 12.2 / 52.1% | 9.7 / 29.3% |
| ZeroBench main (оригинал / с Python) | 17.0 / 34.0% | 9.0 / 12.0% | 9.0 / 17.0% | 12.0% / – | 11.0 / 12.0% |
| MuirBench | 77.5% | 72.3% | 77.6% | 65.9% | 72.7% |
| BVB | 68.4% | 47.0% | 50.5% | 70.7% | 52.1% |
| Рассуждения и знания | |||||
| HLE-Diamond | 15.2% | 7.2% | 9.1% | 15.3% | 8.8% |
| CritPt v2 (Full / Hard) | 31.4 / 17.0% | – / 4.0% | – / 7.0% | 10.6 / 4.0% | 36.9 / 13.5% |
| Apex 2025 | 34.4% | 5.2% | 4.2% | 18.8% | 22.9% |
| ARC-AGI (v1 / v2) | 84.5 / 49.3% | – | – | 87.5 / 42.4% | 22.7 / 1.5% |
| LongBench v2 | 66.9% | 52.9% | 33.1% | 52.3% | 62.1% |
| GraphWalks BFS 256K | 69.7% | – | – | 50.4% | 63.3% |
| MRCR v2 (8-needle, 256K–512K) | 46.4% | – | – | 38.0% | 5.7% |
| AA-Omniscience v1.1 (score) | -8.6 | -48.5 | -32.2 | -6.4 | -2.9 |
| AA-Omniscience v1.1 (non-hallucination rate) | 63.4% | 14.5% | 16.4% | 71.4% | 69.5% |
| ComplexConstraints v1.1 | 33.4% | 31.6% | 35.6% | 30.1% | 30.1% |
| Безопасность и alignment | |||||
| FORTRESS adversarial (Public) | 89.5% | 67.8% | 90.3% | 78.8% | 79.6% |
| FORTRESS benign (Public) | 96.4% | 97.6% | 89.1% | 97.8% | 92.7% |
| StrongREJECT (none) | 97.8% | 98.1% | 99.0% | 99.0% | 99.4% |
| Misbehavior (Internal) ↓ | 9.8% | 27.6% | 46.4% | 57.7% | 52.8% |
| Covert misbehavior (Internal) ↓ | 8.3% | 24.3% | 45.6% | 33.2% | 43.4% |
| Destructive actions (Internal) ↓ | 10.2% | 75.8% | 68.8% | 35.2% | 65.6% |
Модель при пятикратной разнице в активных параметрах (5b против 27b) показывает на AA Index те же 20,1 балла, что и Qwen3.8-27B. В агентском кодинге Inkling часто впереди: SWEBench Verified 79,2% против 77,8%, SWEBench Pro (Hard) 77,9% против 73,8%, DeepSWE 48,2% против 42,2%. Заметный отрыв и в агентном поиске: на BrowseComp 79,0% против 61,6%. Но есть и просадки: Terminal Bench v2.1 (71,6% против 73,0%), MMMU Pro (74,9% против 76,3%), GDPVal-AA (1319 против 1427) и доля ответов без галлюцинаций в AA-Omniscience (63,4% против 71,4%).
Ссылка на веса модели – https://huggingface.co/thinkingmachines/Inkling-2-Mini