The practical middle of the Igno family
Not every workflow needs the heaviest model available. A lot of real work is smaller than that — quick research, a document that needs cleaning up, a task with two or three steps, a question that deserves a solid answer without a long detour through planning and verification.
That's the gap Igno 1 Antarctic is built for.
Antarctic sits between Igno 1 Arctic, our original Igno model, and Igno 1 Pacific, our current flagship. It carries over most of what made Pacific a meaningful jump in reasoning and tool use, but it's tuned to be faster and cheaper to run — a model for the bulk of everyday work rather than the hardest 10% of it.
Where Antarctic fits
Think of the family as a range rather than a ladder. Arctic proved Igno's tool environment could work at all. Pacific pushed as far as we could on reasoning, thread continuity and long, messy multi-tool tasks. Antarctic asks a narrower question: how much of Pacific's judgment can we keep while making the model meaningfully lighter to run?
The answer, so far, is: most of it.
Internal Finkkle evaluation. Thread Continuity measures retention of user intent, constraints, prior decisions and unresolved objectives across multi-turn work sessions.
Antarctic doesn't hold a thread quite as long as Pacific does, but it holds it considerably longer than Arctic, and for most day-to-day conversations that difference never becomes visible.
Tool use that's reliable, not maximal
Antarctic inherits Pacific's approach to tools rather than a scaled-down version of it — it selects the right tool, constructs arguments correctly, and recovers from a failed step instead of stalling. What it doesn't try to do is plan six moves ahead for tasks that don't need it.
Results from Finkkle's internal pre-release evaluation suite. Deterministic and judge-assisted scoring across simulated Igno workflows. Not independently audited.
The gap that matters most in practice is recovery. Antarctic won't handle a badly broken multi-service workflow as gracefully as Pacific will, but it's well past the point where a single unexpected tool result derails the whole task.
Grounding without over-searching
Antarctic uses Finkkle Search the same way Pacific does — as part of reasoning rather than a separate step — and it's noticeably better at this than Arctic was, even if it's not quite as sharp as Pacific at spotting the edge of its own knowledge.
Percentage of internal Grounding Judgment cases where the model correctly determined whether external retrieval was required before answering.
Capability profile
Scores are normalized results from Finkkle's internal evaluation suites and shouldn't be compared directly with third-party benchmark scores.
Pacific WorkBench
On our largest evaluation — messy, multi-stage simulated work with changing instructions, irrelevant information and tools that fail mid-task — Antarctic lands squarely between its siblings.
Hardest multi-stage subset: Arctic 52.9%, Antarctic 68.8%, Pacific 84.6%. A run only counts as successful when the final result satisfies the original objective and every constraint still standing.
Checking its own work
Antarctic revisits assumptions the way Pacific does, just with a shorter leash — it's more likely to catch a contradiction than Arctic was, though Pacific still edges it out on the hardest cases.
Measured on an internal subset of tasks containing hidden contradictions, incomplete evidence or recoverable intermediate mistakes.
Where Antarctic actually wins: speed
This is the point of the model. Antarctic isn't trying to beat Pacific on the hardest 10% of tasks — it's trying to make the other 90% faster and cheaper without giving up much of what made Pacific worth using in the first place.
On lightweight conversational requests, Antarctic cuts median time-to-first-response by roughly 18% compared with Arctic. It's not the 31% improvement Pacific manages, but it comes with capability numbers much closer to Pacific's than to Arctic's — which is exactly the trade-off it's designed around.
Same system, different weight class
Antarctic plugs into the same Igno system as every other model in the family — Finkkle Search, Tasks, Plugins, Remote, Relay, Shield, Projects — with no separate configuration and no different way of talking to it. You don't choose Antarctic by name in most cases; Igno routes lighter, more routine work to it automatically, and reaches for Pacific when a task's complexity calls for it.
For teams running Igno at scale, that matters as much as any benchmark: most requests don't need the biggest model in the family, and Antarctic exists so they don't have to use it.
Benchmark note
All Antarctic benchmark results shown in this article are from Finkkle's internal pre-release evaluation suites, alongside previously published Arctic and Pacific results for comparison. They are intended to measure performance inside representative Igno workflows and have not been independently audited. Internal scores may change as evaluation sets, model versions and the Igno runtime are updated.