SUBJECT Signal Brief: Half-price frontier models and agents that clock in
GPT-6 Sol & Luna
GPT-6 Sol & Luna — launched September 27 — expands the GPT-6 family with two cost-frontier models trained with Astra's methods: API prices drop 50% (Sol to $2/$10 and Luna to $0.10/$0.50 per million tokens) while Sol outscores Claude Opus 5 on business-workflow benchmarks at roughly a tenth of the cost per task.
WHY IT MATTERS Model-cost curves are roadmap inputs — a 50% cut at the frontier tier moves whole feature classes (always-on agents, per-message evals, bulk summarization) from too-expensive to default-on, so the build-vs-wait math on every AI roadmap item needs a re-run this week.
openai.com →Harmony
Harmony — launched September 27 — is an AI-native ITSM platform where employees ask in Slack or Teams and 100+ production-ready agents resolve password resets, app access, onboarding, and device issues end to end, escalating to a human only when needed.
WHY IT MATTERS Internal support is the first agent deployment many B2B companies actually ship — a helpdesk rebuilt around agents rather than a chatbot bolted onto a ticketing tool is the reference architecture PMs will be asked to copy for their own support and ops surfaces.
harmony.io →Cuey
Cuey — launched September 27 — cross-checks your AI's answer against ChatGPT, Claude, Gemini, and 30+ other models in one tab, surfacing where the models disagree and what your model missed, with no copy-pasting between tabs and no extra subscriptions.
WHY IT MATTERS PMs increasingly make calls on model output they cannot independently verify — a one-tab second opinion makes multi-model consensus a lightweight daily eval habit, and signals that showing where models disagree is becoming a user-facing trust pattern worth borrowing.
cuey.io →GoodSocials
GoodSocials — launched September 26 — is an AI social media manager for LinkedIn that drafts authentic posts, runs them through a kanban-style approval board you control, and schedules them so founders and teams post close to daily.
WHY IT MATTERS Founder-led LinkedIn is now a primary B2B distribution channel — the draft-approve-schedule loop is the governance model that makes AI-generated content acceptable to brand-conscious teams, a pattern worth noting for any AI content feature with a human sign-off step.
goodsocials.co →Kaiku
Kaiku — launched September 25 — is a task tracker and wiki for teams whose work is increasingly done by AI agents: it mirrors the API of the tracker most companies already run so existing MCP servers work unchanged, summons agents in comments that propose but never decide, and records every agent run's cost on the issue it worked on.
WHY IT MATTERS When agents do the work, the tracker becomes the system of record for labor — per-issue agent cost accounting and propose-don't-decide permissions are the governance primitives PMs will be expected to spec as agent spend turns into a managed line item.
kaiku.tech →