# Tinker团队发布了Inkling-Small开源模型，性能接近Inkling但体积缩小4倍，支持多模态和长上下文，为开发者提供高效选择

- 来源：Soumith Chintala
- 发布时间：2026-07-31 01:55
- AIWatch 分数：65
- AIWatch 标记：当日精选
- AIWatch 链接：https://aiwatch.icu/events/evt_01kyt2zgmfmj1a52nrw3w3qejg
- 原文链接：https://x.com/soumithchintala/status/2082887820679512573

## 精选理由

常规快讯，保留列表

## AI 摘要

Tinker团队发布了Inkling-Small开源模型，性能接近Inkling但体积缩小4倍，支持多模态和长上下文，为开发者提供高效选择。

## 正文

Today, we are releasing Inkling-Small, an efficient open-weights model that achieves comparable performance to Inkling at a quarter of its size.

Inkling-Small is a Mixture-of-Experts transformer with 276B total parameters, 12B active, trained on NVIDIA GB300 NVL72 systems. Like Inkling, it features native reasoning over audio and images, variable thinking effort, a context window of up to 1M tokens, and well-rounded performance across a range of benchmarks.

Compared to Inkling, Inkling-Small achieves comparable performance with much less compute. Across agentic tool use (Terminal-Bench 2.1), reasoning (HLE text-only, no tools), and instruction following (IFBench) benchmarks, Inkling-Small is more efficient than Inkling and competitive with other models in its weight class. Furthermore, its variable thinking effort lets users easily adapt it to target their use case, balancing cost and performance.

We are releasing the full weights of Inkling-Small. We’re also making it available for fine-tuning on Tinker, and for text, image, and audio chat on Tinker Playground.

Capabilities

As we build our model family, we are always iterating on our approach. Inkling-Small began training after its larger counterpart, which let us improve its training process. For example, we made changes to Inkling-Small’s pre-training data mix and machine learning recipe. Additionally, we post-trained an earlier checkpoint, Inkling-Small (preview), in part using on-policy distillation with Inkling as the teacher. Starting from that checkpoint, we continued scaling agentic coding RL for two weeks. With these improvements, Inkling-Small surpassed Inkling on reasoning and agentic coding benchmarks. Inkling maintains an advantage on knowledge coverage and factuality.

Reasoning and Agentic Tasks

Inkling-Small matches or exceeds Inkling on reasoning and agentic tasks. On Humanity’s Last Exam it scores 31.6%, ahead of Inkling’s 29.7%, and the advantage holds at every thinking budget: Inkling-Small’s test-time compute curves sit above Inkling’s throughout. On SWEBench-Verified it exceeds 80%.

Across many reasoning and agentic benchmarks, Inkling-Small at max reasoning effort has a strong performance-token tradeoff when compared to open weights models in its weight class.

Inkling-Small also runs smoothly across a variety of coding and agent harnesses, making it a cost-efficient choice for coding and tool-use workflows.

Multimodality

Like Inkling, we crafted Inkling-Small for audio intelligence, making it a good candidate for real-world audio applications. We also improved its ability to use Python for visual tasks. The model can combine visual reasoning with operations such as cropping, zooming, and programmatic image inspection, improving usability on documents and charts where relevant information may be small or difficult to inspect directly.

Inkling-Small uses the same natively multimodal encoder-free architecture as Inkling. Audio is represented as dMel spectrograms, while images are divided into 40×40-pixel patches and transformed using a four-layer hMLP. Both are transformed via a light-weight embedding layer and processed jointly with text tokens. Inkling-Small nearly matches Inkling across most multimodal evaluations at a lower cost. It retains strong performance on visual reasoning, chart and diagram understanding, mathematical visual question answering, speech understanding, and longer-form audio reasoning.

Open weights	Closed weights
Inkling-Small	MiMo V2.5	Nemotron-3Nano-Omni	Qwen3-Omni	Qwen3.5397B-A17B	Qwen3.5Omni-Plus	Gemini 3.5Flash-Lite
Vision
MMMU Pro(Standard 10)	74.0%	75.4%	53.0%	60.0%	77.3%	71.0%	79.0%
Charxiv RQoriginal / with python	77.4/81.3%	81.0%/–	63.6%/–	61.1%/–	80.8%/–	72.5%/–	70.0%/–
Audio
Audio MC	54.9%	30.4%	23.2%	24.3%	–	37.6%	33.6%
MMAU	77.0%	73.6%	76.7%	77.5%	–	81.1%	75.2%
VoiceBench	90.1%	86.4%	89.4%	88.8%	–	92.4%	85.9%

Audio and vision benchmarks against specialist omni models (open- and closed-weight), reported at effort=0.99.

Epistemics

Inkling-Small was trained similarly to Inkling on epistemics, focusing on calibration, instruction following, and resistance to censorship. Inkling-Small matches Inkling’s performance on forecasting. Calibration involved RL against proper scoring rules on a large corpus of real-world forecasting questions, improving its ability to express appropriate confidence and produce calibrated forecasts under uncertainty.

Open weights	Closed weights
Inkling-Small	Inkling	Kimi K2.6	GPT-5.5	Claude Opus 4.8	Gemini 3.1 Pro	Grok 4.3
Forecasting
ForecastBenchno search · Brier Index ↑	61.3 ± 0.46	60.1 ± 0.54	58.8 ± 0.41	59.3 ± 0.33	56.2 ± 0.71	60.6 ± 0.43	60.9 ± 0.38
ForecastBenchwith search · Brier Index ↑	61.5 ± 0.54	61.0 ± 0.56	–	64.3 ± 0.79	59.9 ± 0.38	61.9 ± 0.69	61.3 ± 0.54
Prophet ArenaBrier Score ↓	0.1238 ± 0.0086	0.1276 ± 0.0092	0.1265 ± 0.0093	0.1179 ± 0.0089	0.1181 ± 0.0087	0.1155 ± 0.0084	0.1264 ± 0.0089

ForecastBench and Prophet Arena results were obtained during testing between July 19 and July 28, 2026.

Safety

Inkling-Small inherited the same safety post-training recipe as Inkling, with built-in safeguards covering our internal spec of safety. These include everyday human-AI interactions as well as dual-use capabilities. Inkling-Small also underwent the same pre-deployment testing process, comprising both internal evaluations and red-teaming by trusted external partners.

On StrongREJECT, which measures whether models refuse unambiguously harmful requests, Inkling-Small is on par with Inkling, and matches the performance of existing open-weights models. On FORTRESS, which measures safety in settings spanning crime, violence, and dual-use risks, it is competitive in both refusal and over-refusal.

Inkling-Small	Inkling	Nemotron 3Ultra	Qwen3.5397B-A17B	MiMo V2.5	Minimax M2.7	DeepSeek V4Flash
FORTRESSadversarial	71.6%	78.0%	77.6%	77.3%	64.8%	86.3%	32.0%
FORTRESSbenign	96.9%	95.9%	90.6%	95.4%	94.6%	90.1%	99.2%
StrongREJECT	98.4%	98.6%	98.7%	99.4%	99.3%	99.4%	97.4%

Safety benchmarks, reported at effort=0.99; higher is better throughout. FORTRESS adversarial is the rate of refusing harmful requests, benign the rate of still answering safe ones.

Benchmarking Inkling-Small

Like Inkling, we benchmarked Inkling-Small on a broad range of capabilities. All evals run at effort 0.99 and temperature 1.0. All coding evals run with 256K max-token trajectory limit, similar to Inkling.

To improve consistency, we rely on externally reported evaluations for both internal and external models when applicable. Specifically, we use the scores reported by:

Artificial Analysis: Humanity’s Last Exam, GPQA Diamond, SciCode, GDPval-AA v2, Tau 3 Banking, AA Omniscience, MMMU Pro, AA-Briefcase
Scale AI: AudioMC, MCP Atlas
ARC Prize: ARC-AGI v1 and ARC-AGI v2
Forecasting Research Institute: ForecastBench
ProphetArena

Open weights	Closed weights
Inkling-Small	Qwen3.5397B-A17B	MiMo V2.5	Minimax M2.7	DeepSeek V4Flash	Nemotron 3Ultra	Inkling	Claude 4.5Haiku	Gemini 3.5Flash-Lite	GPT 5.6Luna
Model Info
AA Indexv4.1	40.0%	34.0%	37.0%	38.0%	40.0%	38.0%	41.0%	30.0%	36.0%	49.0%
Params (B)activated / total	12 / 276	17 / 397	15 / 310	10 / 230	13 / 284	55 / 550	41 / 975	–	–	–
Agentic (coding)
SWEBench Verified*	80.2%	76.4%	71.0%	79.9%	79.0%	70.7%	77.6%	73.3%	75.0%	93.0%
SWEBench Propublic	55.9%	50.9%	56.1%	56.2%	52.6%	46.4%	54.3%	39.5%	54.2%	62.7%
Terminal Bench 2.1*best harness	64.7%	51.3%	63.7%	55.4%	61.8%	56.4%	63.8%	44.2%	54.0%	82.5%
SciCode	48.7%	42.0%	43.1%	47.0%	44.9%	39.9%	46.1%	43.3%	40.9%	50.0%
Agentic (general)
GDPval-AA v2	1269	962	1145	1159	1189	1164	1238	911	1139	1530
MCP Atlaspublic / all	79.6/79.2%	74.2%/–	–	49.4%/–	69.0%/–	47.4/44.7%	78.8/76.0%	41.2/40.2%	79.8/76.8%	77.0/75.0%
Tau 3 Banking	15.5%	13.4%	6.6%	8.9%	22.9%	13.8%	23.7%	9.1%	16.5%	24.3%
BrowseCompwith context management	77.4%	78.6%	–	76.3%	73.2%	63.0%	77.1%	–	–	84.0%
Toolathlon Verified	54.4%	40.7%	49.1%	47.5%	50.9%	34.3%	45.5%	26.9%	57.1%	67.9%
AA-Briefcase	917	–	–	–	833	870	839	612	–	–
Reasoning (general)
GPQA Diamond	89.5%	89.3%	84.9%	87.4%	89.4%	86.7%	87.2%	67.2%	83.8%	89.5%
HLEtext only	31.6%	27.3%	25.2%	28.1%	32.1%	26.6%	29.7%	9.7%	17.5%	35.6%
HLE†with tools	47.8%	48.3%	40.0%	40.3%	45.1%	37.4%	46.0%	17.8%	42.5%	48.9%
AIME 2026	95.5%	93.3%	93.6%	87.7%	95.8%	94.2%	97.1%	85.1%	82.2%	97.6%
HMMT Feb 2026	90.2%	87.9%	82.6%	71.2%	93.9%	78.8%	86.3%	66.7%	63.6%	98.5%
CritPt	8.3%	1.7%	3.7%	0.6%	7.1%	3.1%	5.4%	0.0%	0.0%	20.6%
Reasoning (abstract)
ARC-AGI-1	84.0%	–	–	–	–	–	79.5%	47.7%	–	87.7%
ARC-AGI-2	40.1%	–	–	–	–	–	36.5%	4.0%	–	47.6%
Factuality
SimpleQA Verified	20.6%	26.0%	16.1%	13.5%	34.1%	32.4%	43.9%	5.9%	44.1%	41.7%
AA Omniscienceindex	-9.0	-29.8	-9.3	0.7	-22.9	-1.0	2.1	-4.2	6.9	-11.6
Chat
IFBench	82.2%	78.8%	67.1%	75.7%	79.2%	81.4%	79.8%	54.3%	78.6%	67.3%
Global-MMLU-Lite	86.7%	90.0%	83.5%	83.9%	88.4%	85.6%	88.7%	83.4%	89.4%	88.7%
Safety
StrongREJECT	98.4%	99.4%	99.3%	99.4%	97.4%	98.7%	98.6%	98.6%	97.6%	98.7%
FORTRESSadversarial	71.6%	77.3%	64.8%	86.3%	32.0%	77.6%	78.0%	91.3%	70.7%	83.8%
FORTRESSbenign	96.9%	95.4%	94.6%	90.1%	99.2%	90.6%	95.9%	94.1%	95.5%	97.8%
Vision
MMMU ProStandard 10	74.0%	77.3%	75.4%	–	–	–	73.5%	58.6%	79.0%	78.6%
Charxiv RQoriginal / with python	77.4/81.3%	80.8%/–	81.0%/–	–	–	–	78.1/82.0%	57.4%/–	70.0%/–	81.4%/–
Audio
Audio MC†	54.9%	–	30.4%	–	–	–	56.6%	–	33.6%	–
MMAU	77.0%	–	73.6%	–	–	–	77.2%	–	75.2%	–
VoiceBench†	90.1%	–	86.4%	–	–	–	91.4%	–	85.9%	–

Inkling-Small against open- and closed-weights models across the full eval suite. Activated and total parameters are given for scale; a dash means the score was not available at the time of writing.*SWEBench Verified: Inkling and Inkling-Small’s numbers are reported using a bash-only harness. We use self-reported numbers for external models.*Terminal Bench 2.1: Inkling and Inkling-Small’s numbers are reported using an internal coding harness. A small number of solutions were found to be contaminated from web search and were assigned a score of 0. We use self-reported numbers for external models where available. Otherwise, we report performance using our internal harness.†Audio MC: Other models were evaluated internally since they are not on the official leaderboard.†VoiceBench: VoiceBench uses rule-based, hard-coded string matching for grading, making the evaluation sensitive to output-formatting differences. We therefore added a system message instructing models to follow the expected answer format.†HLE with tools: We benchmarked Minimax M2.7, Claude 4.5 Haiku, Gemini 3.5 Flash-Lite, and GPT 5.6 Luna using our internal harness.

Try Inkling-Small on Tinker

Tinker customers have seen firsthand that the right fine-tuned model can outperform closed models on a variety of tasks, and do so faster and cheaper. We believe Inkling-Small’s combination of broad performance and efficiency will make it easy to experiment with and test in real applications, and we’re excited to see what developers build with it.

Inkling and Inkling-Small are available on Tinker with a limited-time discount and can be chatted with on the Tinker Playground using text, image, and audio. All other models on Tinker and checkpoints are also now available on Tinker Playground, billed at the rates listed on our pricing page.

Inkling-Small was made in pursuit of our mission to build AI that extends human will and judgment. It’s a capable and efficient model, and an important stepping stone for us as we continue our work.