12 corrections on record · most recent first
-
This piece said Kimi K3's license demands up to a 30% revenue share from large commercial users above $20 million in sales. The license requires a separate agreement from model-as-a-service sellers above $20 million in revenue and names no rate; the up-to-30% figure came from one anonymous Reuters source.
-
This piece said Kimi K3 carries a 30% revenue share above $20 million in sales. Its license requires model-as-a-service sellers above $20 million in revenue to sign a separate agreement with Moonshot; the up-to-30% share was reported by one anonymous Reuters source and is not set in the license terms.
-
This piece gave the unchanged base that GLM-5.3 shares with GLM-5.2 as 743B parameters. The correct figure is 753B.
-
This piece said Kimi K3's license seeks up to a 30% revenue share above $20 million in sales; the license requires model-as-a-service operators above $20 million in twelve-month revenue to sign a separate agreement and names no rate, and the 30% figure came from one anonymous Reuters source. It also said K3 held a green flag for weights already shipped in July, when the weights were promised for and published on July 27. GLM-5.2 was listed under a bespoke license with acceptable-use clauses; it ships under the standard MIT license as a roughly 753B-parameter MoE.
-
This piece said Kimi K3's license seeks a revenue share of up to 30% above a $20 million sales threshold. The license requires model-as-a-service sellers above $20 million in revenue to sign a separate agreement with Moonshot and names no rate; the up-to-30% figure came from one anonymous Reuters source, not from the license terms.
-
This piece said Moonshot's Kimi K3 license seeks up to 30% of revenue above a $20 million threshold. The license requires model-as-a-service sellers above $20 million in revenue to sign a separate agreement with Moonshot and names no rate; the up-to-30% figure came from one anonymous source cited by Reuters, not from the license terms.
-
This piece said China's open-weight fortnight put four near-frontier models into permanent circulation. At publication two of the four were not yet downloadable: Kimi K3's weights were promised for July 27 and Qwen 3.8-Max was a preview.
-
This piece said Kimi K3 shipped genuine open weights on July 16 and graded it green; K3 launched that day on Moonshot's app and API with weights promised by July 27 and no published license, so it is now graded yellow. The piece also said GLM-5.2 carried a bespoke license and an undisclosed parameter count. GLM-5.2 ships under the standard MIT license as a roughly 753B-parameter MoE, and is now graded green.
-
This piece described Kimi K3 as a 2-trillion-parameter checkpoint whose open weights shipped July 16 under a permissive license. K3 has about 2.8 trillion parameters and launched July 16 on Moonshot's app and API with weights promised by July 27; no license text had been published when this piece ran.
-
This piece said Moonshot released Kimi K3's weights on July 16 and called K3 the largest open-weight model released to date. On July 16 Moonshot launched K3 on its app and API and promised the full weights by July 27, 2026; the weights were published on July 27.
-
This piece said the 'Evals are a scam' post appeared four months before the a16z LMArena investment memo. The memo is dated May 28, 2025 and the post September 5, 2025, so the post came just over three months after the memo, not before it.
-
The description said MCP was twelve months old. Anthropic announced the protocol on November 25, 2024, so it was sixteen months old at publication, as the body of the piece already stated.
How we handle errors
When we get something wrong, we correct it in place in the original piece, add a dated note at the top of that piece saying what was wrong and what is correct now, and log the same note here. The note names the error, so the record is checkable.
Three kinds of changes warrant a correction notice:
- Factual errors. Wrong numbers, dates, names, attributions. Anything a reader could verify against a primary source.
- Misleading framings. Statements that were technically true but read as something stronger than the evidence supported. We treat these like factual errors when they materially change a reader’s takeaway.
- Updated facts. A claim was true when filed and isn’t any more. We don’t backdate; we add an Updated note with the new state and the date the fact changed.
Typos and style edits don’t go here. Code-snippet bugs don’t either; those get an inline edit and a Last updated note in the piece itself.
How to flag one
Write to corrections@thedeepfeed.ai. If you’re responding to a piece’s argument rather than a factual error, that goes to letters@thedeepfeed.ai instead.
Corrections policy v1.0 · adopted April 2026