Models & Usage
Most AI-tooling plans meter you tightly because the model underneath is expensive. A plan built on a single first-party frontier API — a Claude Pro seat, say — burns through its allowance fast, because every token is billed at premium first-party rates.
Sigilix is built differently. We run a roster of cheap, highly reliable open and frontier models — the kind that are now industry-standard and dependable for production code work — so our cost base per review is a fraction of a single-vendor stack. That difference is passed straight to you: far more headroom for the same money.
The point isn’t “cheap models, cheaper output.” Every finding still clears the full believability pipeline — evidence, provenance, refute/execute, proof-tier receipts. Lower cost buys you more usage, not lower standards.
The models we run
You can pick the model that fits your work and your usage. The roster includes industry-standard open and frontier models such as:
DeepSeek V4
Strong reasoning for logic and architecture work.
Kimi K2.7
High-quality, high-throughput code reasoning.
MiniMax M3
Reliable general-purpose code understanding.
GLM 5.2
Capable, cost-efficient coverage across roles.
…and more are added as the open/frontier landscape moves. These models are reliable enough to stand behind a believability guarantee, and cheap enough that we can give you usage windows a single-vendor plan can’t match.
On the hosted PR ensemble, each specialist — Logic & Architecture, Security, Performance, Tests, unified by the synthesizer — runs the model tuned to its role, behind cross-provider fallback so one provider’s outage can’t silence a specialist. See The Ensemble.
Pick your own model
You’re not locked to one choice. Select the model that matches the work and the budget you want to spend, and change it as your needs change.
On the CLI and agent surfaces, the default is Sigilix’s own model lineup — Boreas, Pyroeis, Astraeus (and Phanes, coming soon) — served through Sigilix’s managed gateway. On paid tiers you can also bring your own key — connect your own provider key (Anthropic, OpenAI, or another provider) and pay that provider directly. Whatever model you bring is still handed Sigilix’s verified earned-context layer; note, though, that Sigilix’s hosted models are tuned to prioritize that injected context most reliably (see the grounding trade-off on the BYOK page).
Generous usage windows
Because the cost base is low, included usage is generous — meaningfully higher than a comparable OpenAI- or Claude-based plan for the same price.
Per-seat base + metered usage
Each tier is a per-seat base plus metered usage for review volume and agent tokens. No fixed PR windows — you pay for your team plus the work Sigilix actually does.
Pick the model that fits
Run a cheaper model for routine volume, or a heavier one where it matters. The choice is yours, per your usage.
Windows reset every 5 days
Usage windows reset on a 5-day cadence — not weekly — so you spend less time waiting on a limit.
For tier pricing and what’s included at each level, see the pricing page.