G-2.0-Lite — Fast & Compact
AIME 93, MATH-500 97, native vision, long context — strong reasoning in a fast, compact, low-cost package.
G-2.0-Lite is the second generation of our compact, fast model, and the single largest generational improvement in our lineup. It is efficient enough to deliver strong results at low cost and latency, and now achieves benchmark results that rival models many times its size.
AIME 2026 at 93 is the headline number — competition-level mathematical reasoning. MATH-500 at 97 places G-2.0-Lite among the strongest in its weight class. GPQA Diamond at 86 and MMLU-Pro at 85 confirm that this is not a narrow improvement — it is a broad capability upgrade.
Two major new features ship with 2.0. First, the context window expands substantially — long context matching G-2.0-Thinking, enough to process entire codebases, contracts, and research corpora. Second, native vision support arrives for the first time — images, documents, and screenshots processed directly without a separate adapter.
G-2.0-Lite inherits the full privacy architecture of the Guardian platform — encrypted in transit and at rest, with minimal retention by default.
G-2.0-Lite is available now via the Onora App.