Many teams struggle to balance AI performance with budget while avoiding dependence on a single provider. Using a massive model for simple lookups wastes money, and relying on one vendor risks lock‑in, sudden API changes, or regional service limits. Fugu’s new releases address these pain points directly.
Fugu Max is built for the best output per dollar. It routes each request to the cheapest model in its pool that can still solve the task—often a lightweight open‑weights or specialized model like NVIDIA Nemotron. Pricing is $2 per million input tokens and $6 per million output tokens, delivering 40‑60 % lower output costs than Sonnet 5, GPT 5.6 Terra, or Kimi K3. In benchmark tests it achieves the top overall score on six suites and expands the cost‑performance Pareto frontier on seven of ten, giving elite‑model quality at 2‑6× lower expense.
Fugu Ultra v2 targets the hardest, multi‑step workloads—complex reasoning, visual data interpretation, and full‑stack software engineering. It scores 48.3 on Chartography (visual reasoning) and 74.3 on DeepSWE (real‑world coding), outperforming models priced three to five times higher per token. These results come without Fable 5, Fable 5.1, or GPT‑6‑Astra in its agent pool, showing that strong capability does not require proprietary heavyweights. The model’s training cutoff is August 28 2026, reducing exposure to future vendor‑specific changes.
Both versions are available today as hosted APIs with an OpenAI‑compatible interface, requiring only a one‑line switch for existing users. There are no open weights to self‑host, and the service is not offered in the EU/EEA, which is clearly stated up front.
Use Fugu Max when cost efficiency is paramount; choose Fugu Ultra v2 when you need peak capability on demanding tasks. The orchestration layer handles model selection automatically, letting you focus on outcomes rather than infrastructure.
#AI #MachineLearning #LLM #CostOptimization #API #ProductLaunch