BUILD

Models & T1 profiles

You don’t pick a raw model name. You pick a T1 profile, a routing strategy that continuously selects the best available model, tools and reasoning depth for your task.

The four profiles

Profile	Best for	Character
`t1-auto`	You want the router to pick per prompt (default in editors)	Heuristic routing across Core, Flow, Max
`t1-core`	Classification, extraction, support, search, high-volume agents	Fast, efficient, grounded
`t1-flow`	Connected tools, teams, long-running creative work	Adaptive orchestration
`t1-max`	Research, code, strategy, consequential or multimodal work	Everything the others do, plus extended thinking, video and audio

Capability matrix

T1 is a strict superset up the ladder: Max does everything Flow can, Flow does everything Core can. If a modality is not listed on a lower tier, use Max (or Auto, which routes there for you).

Capability	Core	Flow	Max
Text in / text out	Yes	Yes	Yes
Tool use (function calling)	Yes	Yes	Yes
Streaming	Yes	Yes	Yes
Vision (images)	Yes	Yes	Yes
PDF / document input	Yes	Yes	Yes
Prompt caching	Yes	Yes	Yes
Extended thinking	No	No	Yes
Video input	No	No	Yes
Audio input	No	No	Yes

Auto routes for you. Send video or audio on t1-auto and the router hard-floors at Max. There is no configuration to change.

Reliability by design

Every tier has a primary and a failover chain across independent providers. When the primary hiccups, the request moves to the next entry in the chain without changing your call. Providers are chosen for uncorrelated failure modes so a bad afternoon at one vendor never stops the tier.

Two-stage routing

Picking a tier is stage one. Stage two picks the right specialist inside the tier for the shape of your request. A code prompt on Flow does not always run the tier default. A long-context prompt on Max does not always run the tier default. The router inspects the prompt and swaps to a specialist when there is a clear better fit at the tier's quality bar.

Task shape	What we optimise for
Code (fenced blocks or coding cue)	Code-tuned specialist within the tier
Structured output (`json_schema`)	Strict schema compliance
Long context (over 50k characters)	Long-context specialist
Autocomplete shape (tiny `maxTokens`, no tools)	Low-latency P50
Everything else	The tier default

The tier promise is preserved. The intra-tier router never crosses tier boundaries: a Flow specialist is always at Flow quality or above, never a Core-tier model. Max stays a strict superset of Flow, and Flow stays a strict superset of Core.

Selecting a profile

// Let the router pick
await taskclan.run({ profile: "t1-auto", goal: "Draft the release notes" });

// Fast + cheap
await taskclan.run({ profile: "t1-core", goal: "Classify this ticket" });

// Balanced default
await taskclan.run({ profile: "t1-flow", goal: "Plan the migration and list open questions" });

// Deep reasoning + everything modality
await taskclan.run({
  profile: "t1-max",
  goal: "Summarise the key moments in this meeting",
  videos: [{ url: "https://example.com/standup.mp4" }],
});

Why profiles, not model names? The frontier moves constantly. Profiles let us route to the strongest model available for your task, and improve it over time, without you changing a line of code.

Model independence

Behind T1, requests may be served by leading open and proprietary models. Providers process your request under terms that restrict training on your data. We never train foundation models on your content without an explicit opt-in, see the AI Policy.

Controlling cost & latency

Default to t1-core for anything high-volume or latency-sensitive.
Reserve t1-max for tasks where quality clearly justifies the cost.
Set maxTokens and timeouts to bound spend on open-ended goals.