Which Task Do I Use?
Find the catalogue task for the job you have (extract, score, summarise, classify, write, translate) and the benchmark page that shows how models do on it.
On Fronset you don’t pick a model. You put a task in the model field of the
request, and we pick a model that is good enough for that task. (The API also
calls a task a capability, in route and field names.)
This page lists the catalogue tasks you can call today, grouped by category
(a group of tasks that need the same kind of model skill). Find the job you have,
then put the task name in model:
client.chat.completions.create(
model="content-set-relevance-scoring", # the task, not an LLM
messages=[{"role": "user", "content": "..."}],
)
Each task name links to its benchmark page. That page shows which models clear the bar on the task, what each one costs to run it, and the prompt templates the task was measured with. Read those prompts to see what input the task expects.
The list matches the benchmark snapshot of 15 September 2026. A task that is on
the benchmark but can’t be called is left out. GET /v1/models with your key
always gives the current list.
Can’t find your job? You can define your own task, with your own prompts, output schema and validators. It runs on your own provider keys; see defining your own task . This is also the way to get fields back in a JSON shape you choose (invoice fields, form data): catalogue tasks have predefined output schemas and validations.
Structured Data & Fact Extraction
Getting precise fields and facts out of text and keeping to a strict output shape.
| If you need to… | Task (links to its details) |
|---|---|
| Pull self-contained factual statements out of a document, optionally tagged with your categories | atomic-fact-claim-extraction |
| Get the section outline (the table of contents) of a long document such as a contract, RFP or specification | profiled-document-structure-extraction |
| Choose which regions matter when researching a company or subject, from a list of regions you give it | research-region-identification |
| Recover data from text that ignored the format you asked for. It returns the result as a single markdown field. | structured_output_extraction |
Content Summarization & Synthesis
Shortening long input without losing the details that matter.
| If you need to… | Task (links to its details) |
|---|---|
| Summarise a document, web page or transcript in full, optionally around a structure you give it. It also strips page clutter such as navigation, ads and cookie notices | structured-content-summarization |
| Write an executive summary of a long report, to a length you set | report-executive-summary-generation |
| Turn source material into a set number of short, checkable claims | evidence-grounded-claim-generation |
| Fetch one web page whose URL you already know and extract its content and metadata. It does not search | known-url-content-extraction |
| Write title and subtitle options for an article or report, pick the best one, and write its SEO title | publication-title-package-generation |
Long-form Content Generation
Writing longer text that stays coherent and consistent in voice.
| If you need to… | Task (links to its details) |
|---|---|
| Write analysis from a set of numbered claims and keep every reference to them intact | reference-preserving-analytical-writing |
| Outline a report: ordered sections, each with a code and what it must cover | report-outline-generation |
| Write the opening of a newsletter issue from a summary and a list of articles | newsletter-copy-generation |
| Design the sections of a publication or knowledge base | publication-section-taxonomy-generation |
| Write a reusable voice and personality description for an author or assistant | voice-profile-generation |
| Produce an accessible colour and typography theme from references and design constraints | visual-theme-configuration-generation |
Social & Promotional Content
Short posts written to each platform’s conventions and length limits.
| If you need to… | Task (links to its details) |
|---|---|
| Write one social post promoting a piece of content, to your channel’s rules, length and link format | promotional-message-generation |
| Plan a series of posts before and after a release | promotional-campaign-generation |
| Decide which communities a piece of content suits, and draft a post for each one that does | community-content-promotion-generation |
| Draft a short public reply to a message, within the facts you allow | public-response-generation |
Relevance, Classification & Matching
Deciding whether something belongs in a bucket or matches a target.
| If you need to… | Task (links to its details) |
|---|---|
| Score a set of documents or search results for relevance to a query, with a score and a reason for each | content-set-relevance-scoring |
| Score a batch of short social posts against a topic | short-post-batch-relevance-scoring |
| Score drafts or summaries you generated against a target specification | generated-content-relevance-scoring |
| Detect the language of a text | language-identification |
| Label a piece of content with one to three subject domains, reusing your existing labels where they fit | content-domain-suggestion |
| Map topics to your taxonomy, reusing existing entries and proposing a new one only when nothing fits | topic-taxonomy-matching |
| Decide whether an incoming public message needs a reply, how risky it is, and what angle to take | engagement-opportunity-triage |
| Review a drafted reply before it goes out: is it ready, and what is wrong with it | public-engagement-reply-review |
| Check whether a community’s rules allow a post you plan to make | community-policy-evaluation |
| Check a proposed persona against your eligibility policy | persona-policy-eligibility-check |
| Match an item to the best candidates in a pool of profiles (a ticket to an agent, an article to an editor) | profile-pool-matching |
| Pick the best fixed-size set of posts from a list of candidates | social-post-portfolio-selection |
| Recommend authoritative sources to research a subject | authoritative-source-selection |
| Recommend online communities to use as research sources | research-community-selection |
Topic Organization & Clustering
Finding groups in content that has no fixed scheme, then naming and ordering them.
| If you need to… | Task (links to its details) |
|---|---|
| Find the themes across a batch of summaries or claims and assign each item to one | thematic-topic-discovery |
| Name a cluster of related items and describe it in a line | topic-cluster-labeling |
| Assign topics to one or more sections from a list you give it (a ticket theme to product areas, a defect to components) | topic-section-assignment |
| Put a set of topics in a sensible reading order | topic-sequence-optimization |
Infrastructure & Utility
Mechanical work: format conversion, rewriting, translation, prompt handling.
| If you need to… | Task (links to its details) |
|---|---|
| Translate one or more texts and keep IDs, URLs, handles and placeholders untouched | batch-text-translation |
| Restore the blank lines in markdown whose line breaks were stripped, without changing any other character | markdown_newline_repair |
| Rewrite a short factual paragraph for clarity and keep every date, count and qualification | metadata-paragraph-rewriting |
| Check extracted claims: tighten the good ones, drop unsupported or duplicate ones | factual-claim-refinement |
| Write search queries for a subject, for a given search platform and region | research-query-generation |
| Repair a search query so it is valid for a given platform, without changing what it looks for | research-query-validation |
| Turn content into a detailed brief for an image generator (the task writes the brief; it does not make the image) | content-to-image-prompt-generation |
| Adapt a prompt for a different model family and keep its placeholders and output contract | model-specific-prompt-adaptation |
| Write the system and user prompts for one section of a report | section-prompt-generation |
Financial Analysis & Trading Decisions
Reading business and financial documents closely enough to judge risk and opportunity.
| If you need to… | Task (links to its details) |
|---|---|
| Analyse a filing, contract or other document along dimensions you set | profiled-document-analysis |
| Analyse one section of such a document in context | profiled-document-section-analysis |
| Identify catalysts, scenarios, risks and their timing for a company or subject | catalyst-and-scenario-analysis |
| Combine several analyses, including dissenting ones, into one decision with a confidence level | multi-perspective-decision-synthesis |
| Turn a free-text brief about a company or subject into a structured research configuration | subject-configuration-generation |
If two tasks look alike
- Scoring relevance. Use
content-set-relevance-scoringfor documents and search results,short-post-batch-relevance-scoringfor short social posts, andgenerated-content-relevance-scoringfor text your own pipeline produced. - Summarising.
structured-content-summarizationkeeps the detail;report-executive-summary-generationkeeps only what a decision needs, to a length you set. - Classifying.
topic-section-assignmentputs items into a fixed list you give it.topic-taxonomy-matchingdoes the same but may propose a new entry.content-domain-suggestionsuggests labels when you have no list. For single-label classification with your own labels and output shape, define your own task.