Carta has opened Evergreen, a model-agnostic cloud coding platform where teammates and agents share sessions and agents can push pull requests. Chief executive Henry Ward’s 5 November 2026 launch post frames it as collaborative software work, not a single-player chat box. The Evergreen FAQ, updated 4 November 2026, sets the commercial terms: no seat fees, a seven-day trial on a workspace you own, model calls at provider prices, an Evergreen fee of 15 cents ($0.15) per million tokens, and workspace storage at 35 cents ($0.35) per GB-month.
Carta’s own productivity story sits beside that price card. The blog says internal PR volume rose from about 1,300 a week in January to almost 20,000 in the week shown on a chart dated 4 November 2026 — a claimed 15× rise — with more than 6× faster delivery and an 80% lower revert rate. It started with more than 250 engineers and later more than 80% of the company. Those are Carta’s internal operating metrics, not an outside audit of code quality.
Fees: two token lines, not one sticker price
Evergreen’s FAQ separates what you pay the platform from what you pay the model vendor. Model usage passes through at the provider’s price. On top of that, Evergreen charges $0.15 per million tokens for its own fee. Storage is billed at $0.35 per GB-month. Platform fee, agent time, storage and search charges are waived for seven days on a workspace you own; after that, usage-based billing applies. There are no seat licences in the FAQ’s description.
That structure means a bill can move with both model choice (Claude, Codex, Grok, Muse and others Carta names) and how heavy the sessions are. Comparing Evergreen to a flat per-seat IDE without adding the provider pass-through would understate cost; ignoring the $0.15 Evergreen fee would understate it the other way.
Procurement teams should budget for both lines from day eight. A light Claude session and a heavy Codex marathon will share the same Evergreen per-million add-on while the provider line diverges. Storage at $0.35 per GB-month sits beside those token charges, so long-lived workspaces with large repos can move the monthly total even when token burn is quiet. The FAQ is the commercial source of record for those three numbers; TSN is not inventing seat caps, free-token allotments or regional list prices beyond what that page states.
How this sits next to Cursor and Factory posts on TSN
Evergreen’s nearest TSN neighbours are other agent-coding launches, not cap-table software news. Cursor Live Apps put cloud agents on building and hosting web apps. Cursor’s Agent SDK treats agents as testable code projects. Factory’s Droid harness sells agent performance with vendor-owned benchmarks. Evergreen’s pitch is multiplayer sessions with audit logs, scoped access and spending caps — a collaboration and control story more than a single-agent leaderboard.
Throughput is not the same as quality
For engineering managers, Carta’s 1,300-to-20,000 PR chart is a throughput claim. An 80% lower revert rate is Carta’s proxy for fewer bad merges, not a static-analysis grade or customer-production defect rate. “More than 6× faster” needs the same attribution: faster by Carta’s internal stopwatch, on Carta’s own repos and process. The January baseline and the 4 November chart week are Carta’s chosen windows; outsiders cannot re-run the series from the blog alone.
Security-minded buyers will care that Evergreen stresses isolated sessions, scoped access, audit logs and spending caps in the launch note. Those controls answer a different question from PR volume: who can see what, and how hard it is to blow the model budget. The seven-day trial is the only free window the FAQ describes; TSN is not inventing longer free tiers or hard token caps beyond that page.
UK and EU readers get no regional data-residency SKU in the launch post or FAQ. Availability is presented as a Carta cloud product with the fee card above. Until Carta publishes methodology for the PR and revert series, those figures remain company operating metrics — useful for understanding how Carta talks about Evergreen’s effect inside its own walls, not a portable proof of code quality for every adopter.
What to watch next
Three public breadcrumbs would move the file: Carta publishing methodology behind the PR and revert series; a pricing page that restates the $0.15 / $0.35 fees outside the FAQ; or named design-partner case studies that separate Evergreen’s effect from Carta’s own process changes. Until then, Evergreen is a multiplayer AI coding launch with Carta-reported throughput gains and a two-part token bill — provider price plus Evergreen’s per-million fee.
Sources
Primary:
- Carta Evergreen: Introducing Evergreen (Henry Ward, 5 Nov 2026)
- Carta Evergreen FAQ (updated 4 Nov 2026)
Related stories
- Cursor Live Apps: Cloud Agents Build and Host Web Apps
- Cursor’s Agent SDK Turns AI Agents Into Code Projects You Can Test
- Factory Says Droid 1.0 Beats Codex and Claude Code in Its Own Tests
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
