aiminute. ← All AI news
New Models 2026-08-14

An American enterprise vendor's new flagship model is a Chinese open model underneath

An American enterprise vendor's new flagship model is a Chinese open model underneath

Writer, which sells AI to large regulated companies, released Palmyra X6 on 13 August. It is a post-trained variant of GLM-5.2, the open-source model from the Chinese lab Z.ai, rather than something trained from scratch. Alongside it the company shipped a rebuilt agent harness and governance tools aimed squarely at runaway token bills. Writer says its agent product now runs at 52% lower average cost with the new model and harness together, 48% faster, and 10% better on quality; separate work on the harness alone claims a token-spend reduction of close to 40% with no loss of accuracy. The governance side gives IT leaders limits and visibility into where an organisation's tokens are going.

Why it mattersWriter's customers are exactly the buyers usually assumed to insist on an all-American stack. Building its flagship on a Chinese lab's open weights suggests the cost gap has widened enough to outweigh that instinct, and it matches what Chinese outlets have been reporting from the other side for weeks: foreign firms quietly swapping their base layer. The second half matters more to anyone actually running agents. The harness — how many times the model is called, how much context it is handed, when it stops — is where a large share of the bill is decided, and it is now something vendors compete on rather than an implementation detail.
#AI Agents

✓ Verified · 3 sources

WhatsApp X Telegram
Read in the app — free, in 9 languages

Related stories

DeepSeek's cheap workhorse can now see — and on agent tasks that need eyes it says it is close to Anthropic's best
2026-08-21
Given four hours and a GPU to improve the way AI is trained, the best agent scored 0.25 out of 1 — and most never tried
2026-08-21
ChatGPT can now read your iMessages and send them — and the setting that lets it skip asking is the one OpenAI warns about
2026-08-21
Show the robot once — three to twelve seconds — and it gets the job right 59 times out of 100 with no training at all
2026-08-21
The agent invented a second person to vouch for its code. A 24-year-old in Texas refused to believe either of them.
2026-08-21