Kimi K3's Two-Week Flash

The spike

For the week of July 6, Kimi appeared in 2 tagged stories across HN and Lobsters, good for 0.32% share of voice. By July 27 it held 8.05%, the product of 54 stories in a 671-story week. That was enough to surpass Anthropic, which sat at 7.15%, and GPT, at 5.96%. Only Claude (20.72%) and OpenAI (10.73%) ranked higher.

The spike was not manufactured. The HuggingFace release thread reached 1,370 points and 539 comments, placing it among the most engaged HN posts of the period. Four additional Kimi stories cleared 487 points: a SoTA comparison with Fable (871 points, 441 comments), a Kimi Work release (671 points, 275 comments), an architecture overview (499 points, 110 comments), and a 256k-context variant (487 points, 155 comments). These are front-page numbers by any measure.

The reversal was immediate. The week of August 3, Kimi fell to 12 stories and 1.95% share of voice, behind all five tracked models. The partial week of August 10 produced a single story.

No tail

The entity_stats table for Kimi contains 11 snapshots from June 12 through August 10. Every one records job_mentions_7d at 0. This holds before the spike, during it, and after it.

Kimi is a tracked entity in the same job-listing pipeline used for all 65 monitored entities. The absence is not a data gap. It is a signal that the release conversation did not translate into employers naming Kimi in job postings, even at peak buzz.

For comparison, Kimi generated more HN conversation than Anthropic during the July 27 week while producing zero job mentions. Anthropic, a model lab with an established hiring footprint, is the relevant counterfactual: conversation volume alone does not create a hiring pull.

One week only

Kimi's crossover above Anthropic lasted a single week. Before July 27, Kimi never exceeded Anthropic's share of voice. After July 27, it fell behind every tracked model. The ranking for that week was Claude (20.72%), OpenAI (10.73%), Kimi (8.05%), Anthropic (7.15%), GPT (5.96%). By August 3, Kimi was last at 1.95%.

The shape of the curve is the finding. A model that enters the top three by share of voice for one week, then drops to last place the following week, is not accumulating developers. It is consuming a news cycle.