Primary use stated or inferred cautiously from official documentation.
T
THINKING MACHINES LAB / thinkingmachines/Inkling-2-Small
Inkling 2 Small
A multimodal model from Thinking Machines Lab with 594 billion total parameters and 17 billion active per token. It takes text and images and generates text, without audio input. Since November 5, it has been available for sampling and fine-tuning on Tinker and testing in Tinker Playground.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | thinkingmachines/Inkling-2-Small |
|---|---|
| Model type | Multimodal · Code · Text and reasoning |
| Access model | Paid proprietary |
| License | Tinker Terms of Use |
| Deployment | Hosted API |
| Release | 05 NOV 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text · Image |
| Exit | Text |
| Context window | Up to 1M in the model card; Tinker offers 64K and 256K |
| maximum output | Not published |
| Reasoning | Adjustable effort |
| Published tools | Features |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | LoRA on Tinker |
| Verified platforms | Tinker |
Tool-assisted coding, analysis of visual documents and adaptation to custom tasks. Record the effort setting and Tinker identifier when comparing quality, latency and usage. This is a distinct model from first-generation Inkling-Small.
Weights are not yet downloadable. The model card lists up to 1M context, while Tinker offers 64K and 256K for tuning and sampling, and 256K for serverless inference in beta. Distinguish the model limit from the service limit.
03 / CAPABILITIES
What it can do
Orientation
Tool-assisted coding, analysis of visual documents and adaptation to custom tasks. Record the effort setting and Tinker identifier when comparing quality, latency and usage. This is a distinct model from first-generation Inkling-Small.
Context
Up to 1M in the model card; Tinker offers 64K and 256K · Not published
Tools and integration
Features
Access
Tinker · inference and fine-tuning
Input modalities
TextImage
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | 0.30 USD / 1M input tokens | Standard API rate |
| Cached input | 0.03 USD / 1M cached input tokens | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | 1 USD / 1M output tokens | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
Tinker charges USD 0.30 input, USD 0.03 cached input and USD 1 output per million tokens. Training is separate at USD 1.50 with 64K or USD 3 with 256K. Serverless inference has a separate identifier and beta access; confirm the mode before budgeting.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
The card reports 30.6 on Artificial Analysis Intelligence Index v5.0, 72.1% on DeepSWE v1.1 and 86.3% on SWE-bench Pro v2 Hard. Versions and subsets are preserved. Results in the provider card do not establish performance under different settings or reproduction by Inferama.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| Artificial Analysis Intelligence Index v5.0 | 30,6 | Score | View source ↗ |
| DeepSWE v1.1 | 72,1 % | Score | View source ↗ |
| SWE-bench Pro v2.0 Hard | 86,3 % | Score | View source ↗ |
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
Inkling 2 Small
Version added to Inferama's verified catalog.
07 / SOURCES