Explore AI models by supply chain capability. Understand what each one can help with, the data it needs, and where its limits are before you build.
Model directory
365 entries
Frontier, specialized and open source models with the supply chain capabilities they support. Open a row for its full details. Source snapshot 2026-09-12.
All entries365
Frontier models130
Specialized models158
Open source models77
Showing 1–10 of 365
Model
Group
Supply chain capabilities
Input
Output
Latency
Released
Fugu MaxSakana AI
Frontier models
S&OP and capacity planningProcurement and suppliers+2
$2.00per 1M tokens
$6.00per 1M tokens
4.1 s
Fugu Ultra v2Sakana AI
Frontier models
S&OP and capacity planningProcurement and suppliers+2
$5.00per 1M tokens
$30.00per 1M tokens
4.5 s
DeepSeek V4.1 FlashDeepSeek
Open source modelsMIT
Procurement and suppliersOrders and customer service+1
$0.30per 1M tokens
$1.20per 1M tokens
0.5–8.2 s
Ling 3.0 Flash VLInclusionAI
Frontier models
S&OP and capacity planningProcurement and suppliers+2
$0.00per 1M tokens
$0.00per 1M tokens
0.6 s
Mercury 2.5Inception
Frontier models
S&OP and capacity planningProcurement and suppliers+2
$0.04per 1M tokens
$0.15per 1M tokens
1.5 s
GPT-6 AstraOpenAI
Frontier models
S&OP and capacity planningProcurement and suppliers+2
$10.00per 1M tokens
$50.00per 1M tokens
1.9–4.4 s
GPT-6 Astra (Fast)OpenAI
Frontier models
S&OP and capacity planningProcurement and suppliers+2
$20.00per 1M tokens
$100.00per 1M tokens
—
Gemini 3.8 FlashGoogle
Frontier models
Warehouse operationsTransport and delivery+6
$0.75per 1M tokens
$3.75per 1M tokens
1.2–2.0 s
GLM 5.3 FastZ.ai
Open source modelsOpen weights · Custom model license
S&OP and capacity planningProcurement and suppliers+2
$2.10per 1M tokens
$6.60per 1M tokens
0.8 s
Qwen3.8 Max 0902Alibaba / Qwen
Frontier models
S&OP and capacity planningProcurement and suppliers+2
$2.00per 1M tokens
$6.00per 1M tokens
1.3 s
Prices are base USD rates with the billing unit shown; audio, context, caching and provider tiers can cost differently. Latency is the range of active providers’ median time to first token at the snapshot date. Specifications come from Vercel AI Gateway and linked publisher documentation. Applications are editorial suggestions; these models have not been benchmarked on Upshift data.
Three kinds of model.
Every entry in the directory belongs to one group. The group says how a model is built and delivered, which shapes what it costs, where it runs and the work it does best.
130 entries
Frontier models
The largest general-purpose models from the leading labs, used through a hosted API. They read, reason, write and call tools across almost any task.
Choose them when
The work mixes reading, judgement and writing: extracting fields from messy documents, answering questions over policies and contracts, drafting supplier messages, or planning with your own tools.
Trade-offs
The highest rates per token, and your data leaves your network, so check each provider’s retention terms. New versions replace older ones quickly.
Examples
Fugu Max, Mercury 2.5, Ling 3.0 Flash VL
158 entries
Specialized models
Models built for one kind of input or task: time-series forecasters, image detectors, embeddings and rerankers, speech, and clearly labelled decision methods.
Choose them when
The job is narrow and measurable, such as forecasting demand from history, spotting damage in photos or matching duplicate records. A focused model is often cheaper, faster and easier to test than a general one.
Trade-offs
Each one does a single job, so a workflow may need several. Many must be fitted or validated on your own historical data before you can trust them.
General-purpose models whose weights you can download. Run them on your own servers, or rent them from a hosting provider.
Choose them when
Data must stay in-house, volumes are high enough that per-token rates add up, or you want to fine-tune a model on your own records.
Trade-offs
You run and secure the hardware, and open models usually trail frontier ones on the hardest reasoning. Licences differ: open source permits most uses, while open-weights terms can restrict commercial use.
Examples
DeepSeek V4.1 Flash, GLM 5.3 Fast, Qwen3.8 27B
A useful shortlist starts with context.
Go beyond a model name. Find the information you need to decide what deserves a closer look.
01
Search by capability
Start from a supply chain task, then filter the directory to find relevant models and methods.
02
Understand the trade-offs
Review supported inputs, availability, licenses and published specifications alongside practical uses and limitations.
03
Find a way to begin
Follow the model’s sources and application guides, or explore supporting libraries for implementation.
Put it to work
A place to start.
Specifications and prices are dated source snapshots. The directory is a reference; models have not been benchmarked on Upshift data.
Name the task
Begin with a concrete question: forecast demand, extract information, or support a planning decision.
Explore the options
Use the directory filters and open a model profile to check its fit, data requirements and limitations.
Test on your own example
Follow the linked documentation and compare the result with a baseline you understand.