# Inkling 2 Mini

Thinking Machines Lab · Open Weight · rank 70 · bench-align-v5

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelchoice.dev/models/inkling-2-mini  
JSON: https://modelchoice.dev/api/model/inkling-2-mini

## Facts

| Field | Value | Evidence |
| --- | --- | --- |
| Canonical id | `inkling-2-mini` | — |
| Overall score | 55.9 | Observed 2026-11-21 · source benchlm:models |
| Verified score (BenchLM): a separate BenchLM figure, not the overall score | 67 | Observed 2026-11-21 · source benchlm:models |
| Context window | 1M tokens | Observed 2026-11-21 · source benchlm:models |
| Modalities (models.dev) | Input: text, image · Output: text | Observed 2026-11-21 · source openrouter:pricing |
| Release date | 2026-11-20 | Observed 2026-11-21 · source benchlm:models |
| Access type | Open Weight | — |
| Blended $/1M (75% input / 25% output) | $0.24 | Derived from the input and output rates below |

## Capability evidence

Seven axes from the ranking source. An axis the source did not score is unavailable, not zero. Beside each score is BenchLM's separate verified score for that axis; the two are published side by side and are not interchangeable.

| Axis | Score | Evidence | Verified (BenchLM) | Verified evidence |
| --- | --- | --- | --- | --- |
| Agentic | 65.3 | Observed 2026-11-21 · source benchlm:models | 63.4 | Observed 2026-11-21 · source benchlm:models |
| Coding | 64.6 | Observed 2026-11-21 · source benchlm:models | 64.6 | Observed 2026-11-21 · source benchlm:models |
| Knowledge | 54.4 | Observed 2026-11-21 · source benchlm:models | 54.4 | Observed 2026-11-21 · source benchlm:models |
| Reasoning | 75.3 | Observed 2026-11-21 · source benchlm:models | 75.3 | Observed 2026-11-21 · source benchlm:models |
| Multimodal & Grounded | 73.6 | Observed 2026-11-21 · source benchlm:models | 73.6 | Observed 2026-11-21 · source benchlm:models |
| Instruction Following | Unavailable | Unavailable · source benchlm:models | Unavailable | Unavailable · source benchlm:models |
| Math | Unavailable | Unavailable · source benchlm:models | Unavailable | Unavailable · source benchlm:models |

## Runtime service evidence

Measured values with the date they were observed. Nothing is inferred from a sibling model or a provider claim. Regional or per-endpoint measurements appear only when the API supplies them; none are modelled.

| Measurement | Value | Observed | Last good | Evidence |
| --- | --- | --- | --- | --- |
| Time to first token | Unavailable | Unavailable | Unavailable | Unavailable · source benchlm:speed |
| Throughput | 48 tok/s | 2026-11-21 | 2026-11-21 | Observed 2026-11-21 · source openrouter:endpoints |

## Endpoint and price matrix

Every published price component, including cache reads and writes.

| Component | USD | Evidence |
| --- | --- | --- |
| Input / 1M tokens | $0.15 | Observed 2026-11-21 · source benchlm:pricing |
| Output / 1M tokens | $0.50 | Observed 2026-11-21 · source benchlm:pricing |
| Cache read / 1M tokens | $0.015 | Observed 2026-11-21 · source benchlm:pricing |
| Cache write / 1M tokens | Unavailable | Unavailable · source openrouter:pricing |
| Blended / 1M (75% input / 25% output) | $0.24 | Derived — from the input and output rates above; it has no source record of its own |
| Cost per successful task (LiveBench) | Unavailable | Unavailable · source livebench:table |

**Pricing note.** Thinking Machines Lab’s Tinker models and pricing page lists Inkling 2 Mini at $0.15 prefill / $0.015 cached prefill / $0.50 sample per million tokens on both the 64K training tier (thinkingmachines/Inkling-2-Mini, $0.75 training) and the 256K training tier (thinkingmachines/Inkling-2-Mini:peft:262144, $1.50 training), with 1M sampling context on both options. Open weights (~20GB NVFP4/MXFP8) are also available under Apache 2.0 on Hugging Face for local and self-hosted inference.

## Workload-aware monthly cost example

10 conversations per day × 8 messages × 22 active days, 1200 input and 400 output tokens per message, no cache. Derived here from the published rates above by this site's own calculator — not a figure any source published.

| Field | Value |
| --- | --- |
| Modelled monthly cost | $0.67 |
| Modelled tokens | 2.82M |

## Benchmark record

50 matched benchmark rows with their published value, unit, and provenance.

### Knowledge

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| GPQA (Graduate-Level Google-Proof Q&A) | 86.6 | 448 questions | Graduate level | [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://arxiv.org/abs/2311.12022) |
| GPQA-D (GPQA Diamond) | 86.6 | Graduate-level science questions | Graduate level | [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://arxiv.org/abs/2311.12022) |
| Artificial Analysis Intelligence Index | 20.1 | Cross-benchmark intelligence index | Display-only external reference | [Artificial Analysis](https://artificialanalysis.ai/) |
| AA-Omniscience Index (Artificial Analysis Omniscience Index) | -8.6 | Knowledge questions | Broad factual knowledge | [AA-Omniscience: Knowledge and Hallucination Benchmark](https://artificialanalysis.ai/evaluations/omniscience) |
| AA-Omniscience Accuracy (Artificial Analysis Omniscience Accuracy) | 20.5 | Knowledge questions | Broad knowledge | [Artificial Analysis model benchmarks](https://artificialanalysis.ai/models/grok-4-3) |
| AA-Omniscience Hallucination Rate (Artificial Analysis Omniscience Hallucination Rate) | 36.6 | Knowledge questions | Factuality | [Artificial Analysis model benchmarks](https://artificialanalysis.ai/models/grok-4-3) |
| HLE w/o tools (Humanity's Last Exam without tools) | 30.7 | Expert-level questions | Frontier expert level | [Introducing GPT-5.4 mini and nano](https://openai.com/index/introducing-gpt-5-4-mini-and-nano/) |

### Coding

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| Terminal-Bench 2.1 | 71.6 | Terminal-based software-agent tasks | Professional software engineering | [Terminal-Bench](https://www.tbench.ai/) |
| SWE-bench Verified (Software Engineering Benchmark Verified) | 79.2 | 500 verified issues | Professional software engineering | [SWE-bench: Can Language Models Resolve Real-World GitHub Issues?](https://arxiv.org/abs/2310.06770) |
| SWE Multilingual | 73.7 | Multilingual software-engineering tasks | Professional software engineering | [MiniMax M2.7: Early Echoes of Self-Evolution](https://www.minimax.io/news/minimax-m27-en) |
| DeepSWE | 48.2 | 113 software engineering tasks across 91 repositories and 5 languages | Long-horizon software engineering | [DeepSWE benchmark blog](https://deepswe.datacurve.ai/blog) |
| FrontierCode 2.0 | 15.1 | 132 private software-engineering tasks | Frontier coding-agent quality | [FrontierCode 2.0](https://cognition.com/blog/frontier-code-2.0) |

### Mathematics

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| Apex | 34.4 | Advanced mathematical reasoning | Frontier math reasoning | [DeepSeek-V4 Technical Report](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf) |

### Reasoning

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| LongBench v2 | 66.9 | Long-context tasks | Hard long-context | [LongBench v2](https://arxiv.org/abs/2412.15204) |
| MRCR v2 256K-512K (OpenAI MRCR v2 8-needle 256K-512K) | 46.4 | 100 eight-needle retrieval examples | Very long-context retrieval | [Muse Spark 1.3 Evaluation Methodology](https://research.meta.ai/static/muse-spark-1-3-multimodal-evaluation-methodology) |

### Agentic

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| Terminal-Bench 4.0 | 5.60 | 66 professional computer-work tasks | Frontier autonomous knowledge work | [Terminal-Bench 4.0](https://www.tbench.ai/news/terminal-bench-4-0) |
| Terminal-Bench-Science 0.1 | 1.4 | 70 expert-curated scientific research workflows | Frontier scientific research workflows | [Terminal-Bench-Science 0.1](https://www.terminal-bench-science.ai/) |
| Terminal-Bench 2.1 | 71.6 | Terminal-based software-agent tasks | Professional software engineering | [Terminal-Bench](https://www.tbench.ai/) |
| BrowseComp | 79 | Research questions requiring browsing | Hard web research | [BrowseComp](https://openai.com/index/browsecomp/) |
| AA-AnalystAgent (Artificial Analysis AnalystAgent) | 46.3 | Spreadsheet and document analysis questions | Business and data analysis | [AA-AnalystAgent Benchmark Leaderboard](https://artificialanalysis.ai/evaluations/aa-analyst-agent) |
| JobBench | 49.6 | 130 tasks across 35 occupations | Professional multi-source workflows | [JobBench: Aligning Agent Work With Human Will](https://arxiv.org/abs/2605.26329) |
| MCP Atlas | 82.7 | Tool-integrated agent tasks | Advanced tool use | [Introducing GPT-5.4 mini and nano](https://openai.com/index/introducing-gpt-5-4-mini-and-nano/) |
| DeepSearchQA | 85.3 | Agentic browsing and list-answer questions | Agentic web research | [Muse Spark Eval Methodology](https://ai.meta.com/static-resource/muse-spark-eval-methodology) |
| EnterpriseOps-Gym (EnterpriseOps-Gym (provider runs)) | 42.00 | Enterprise workflow tasks | Stateful enterprise tool use | [EnterpriseOps-Gym: Environments and Evaluations for Stateful Agentic Planning and Tool Use in Enterprise Settings](https://arxiv.org/abs/2603.13594) |

### Multimodal & Grounded

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| Chartography (no tools) (Chartography without tools) | 22.1 | 100 specialized chart types | Professional chart reasoning | [Chartography](https://surgehq.ai/blog/chartography) |
| Chartography (tools) (Chartography with image and code tools) | 72.5 | 100 specialized chart types | Professional chart reasoning | [Chartography](https://surgehq.ai/blog/chartography) |
| MMMU-Pro (Massive Multi-discipline Multimodal Understanding Pro) | 74.9 | Multimodal academic reasoning | Frontier multimodal | [MMMU-Pro: A More Robust Multi-discipline Multimodal Understanding Benchmark](https://arxiv.org/abs/2409.02813) |
| BabyVision w/ Python (BabyVision with Python) | 88.8 | Visual perception tasks with Python | Fine-grained visual perception | [Kimi K3: Open Frontier Intelligence](https://www.kimi.com/blog/kimi-k3) |
| ZeroBench w/ Python (ZeroBench_main with Python) | 34.0 | Visual reasoning questions with Python | Tool-augmented visual reasoning | [Kimi K3: Open Frontier Intelligence](https://www.kimi.com/blog/kimi-k3) |
| PerceptionBench (PerceptionBench (Internal)) | 57.5 | Internal atomic visual-perception tasks | Fine-grained visual perception | [Kimi K3: Open Frontier Intelligence](https://www.kimi.com/blog/kimi-k3) |
| MathVision | 89.0 | Visually grounded math problems | Advanced multimodal mathematics | [Qwen3.6 launch benchmarks](https://qwen.ai/blog?id=qwen3.6) |
| ScreenSpot Pro | 76.5 | 1,581 grounding instructions | Professional GUI grounding | [ScreenSpot-Pro: GUI Grounding for Professional High-Resolution Computer Use](https://arxiv.org/abs/2504.07981) |
| ZeroBench | 17.0 | 100 visual reasoning questions | Tool-augmented visual reasoning | [Muse Spark Eval Methodology](https://ai.meta.com/static-resource/muse-spark-eval-methodology) |
| CharXiv (CharXiv Reasoning) | 88.3 | Scientific chart reasoning | Scientific visualization reasoning | [CharXiv: Charting Gaps in Realistic Chart Understanding in Multimodal LLMs](https://charxiv.github.io/) |
| CharXiv w/o tools (CharXiv Reasoning without tools) | 80.6 | Scientific chart reasoning (tool-free) | Scientific visualization reasoning | [CharXiv: Charting Gaps in Realistic Chart Understanding in Multimodal LLMs](https://charxiv.github.io/) |
| BabyVision | 53.4 | Visual perception tasks | Fine-grained visual perception | [Muse Spark 1.1 Evaluation Report](https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report) |
| OmniDocBench 1.6 (OmniDocBench v1.6) | 92.8 | Document parsing and extraction | Grounded document reasoning | [OmniDocBench: Benchmarking Diverse PDF Document Parsing with Comprehensive Annotations](https://arxiv.org/abs/2412.07626) |

### Instruction Following

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| MultiChallenge (provider reports) (MultiChallenge: provider-reported runs) | 50.00 | — | — | [MultiChallenge](https://labs.scale.com/leaderboard/multichallenge) |

### external

| Benchmark | Value | Tasks | Difficulty | Provenance |
| --- | --- | --- | --- | --- |
| PMPP-Hard (PMPP-Hard GPU kernel optimization (AUC, 0-10 hours)) | 455.82 | 69 hard GPU kernel-optimization tasks | Long-horizon GPU kernel optimization | [Introducing Inkling 2 Small](https://thinkingmachines.ai/news/inkling-2-small/) |
| ScreenSpot Pro w/ Python (ScreenSpot Pro with Python) | 87.7 | 1,581 GUI-grounding instructions on professional screenshots | High-resolution GUI grounding with tools | [Introducing Inkling 2 Small](https://thinkingmachines.ai/news/inkling-2-small/) |
| ComplexConstraints (ComplexConstraints (Surge)) | 33.4 | Prompts with multiple interacting constraints | Constrained instruction following | [Introducing Inkling 2 Small](https://thinkingmachines.ai/news/inkling-2-small/) |
| Diligence Bench (DiligenceBench (Thoughtful), best harness) | 48.6 | Financial due-diligence research tasks | Agentic financial analysis | [Introducing Inkling 2 Small](https://thinkingmachines.ai/news/inkling-2-small/) |
| StrongREJECT (StrongREJECT (no jailbreak)) | 97.8 | Forbidden requests across harm categories | Harmful-request refusal | [A StrongREJECT for Empty Jailbreaks](https://arxiv.org/abs/2402.10260) |
| BVB (BVB (Vision Interpretation and Creation)) | 68.4 | Vision interpretation and creation tasks (as reported by Thinking Machines Lab) | Visual interpretation and creation | [Introducing Inkling 2 Mini](https://thinkingmachines.ai/news/inkling-2-mini/) |
| Intelligence per Watt (Intelligence per Watt (local AI energy efficiency)) | 0.703 | Local LLM query efficiency benchmark suite | Energy-efficient local inference | [Intelligence per Watt: Measuring Intelligence Efficiency of Local AI](https://arxiv.org/abs/2511.07885) |
| CritPt v2 (Full) (CritPt v2 Full set (research-level physics)) | 31.4 | Composite unpublished research-level physics challenges (Full set) | Frontier research-level physics reasoning | [Introducing Inkling 2 Mini](https://thinkingmachines.ai/news/inkling-2-mini/) |
| CritPt v2 (Hard) (CritPt v2 Hard subset (research-level physics)) | 17.0 | Hard subset of CritPt v2 physics challenges | Hard frontier physics reasoning | [Introducing Inkling 2 Mini](https://thinkingmachines.ai/news/inkling-2-mini/) |
| GraphWalks BFS 256K (GraphWalks BFS at 256K context) | 69.7 | Multi-hop BFS graph traversal problems at 256K context | Long-context multi-hop graph reasoning | [Introducing Inkling 2 Mini](https://thinkingmachines.ai/news/inkling-2-mini/) |
| FORTRESS (adversarial, Public) (FORTRESS Public adversarial safety split) | 89.5 | Public adversarial safety prompts from FORTRESS | Adversarial safety and refusal | [Introducing Inkling 2 Mini](https://thinkingmachines.ai/news/inkling-2-mini/) |
| FORTRESS (benign, Public) (FORTRESS Public benign over-refusal split) | 96.4 | Public benign safety-adjacent prompts from FORTRESS | Calibration against over-refusal | [Introducing Inkling 2 Mini](https://thinkingmachines.ai/news/inkling-2-mini/) |

## Lifecycle and limitations log

Lifecycle events the source associates with this model.

No lifecycle event references this model. That is not evidence the model has no lifecycle plan — only that this source published none.

## What this profile does not claim

Values are reproduced exactly as their sources published them, in the units those sources declared; none are converted, interpolated, or averaged across providers. Any field marked unavailable was attempted and not returned. Last attempted fetch for this model's score: 2026-11-21 05:17 UTC.
