Eigen RadarAI
Archive
Analysis

Mozilla puts the open-weight lag at 4.4 months

State of Open Source AI v1.1 on 15 September puts Moonshot's Kimi K3 three points behind Anthropic's Fable 5 on Artificial Analysis, at about 30 percent of the listed price, and fits the METR horizon gap at about 4.4 months. AI Chat Daily repeats those scores and CTO Raffi Krikorian's closed-model window of eight to 12 hours of expert time.

Artificial Intelligence··Morning
Two contrasting industrial heat-sink blocks sit on a bright laboratory bench: a larger dark metal block in front and a smaller pale ceramic block set farther back across an empty gap.

Kimi K3 sits three points behind Fable 5

Ars Technica says Mozilla's recurring report stresses that open-weight models let anyone download and run the files while training data and code often stay withheld. Moonshot's Kimi K3 sits three points behind Anthropic's closed Fable 5 on the Artificial Analysis Intelligence Index composite and around 30 percent of the listed price. Mozilla fits METR's 50 percent success time horizon at about 4.4 months; closed systems currently lead on 8-to-12-hour jobs, and open systems reach that band four months later. On Vals AI's shared-harness Terminal-Bench 2.1 run, Z.ai's GLM 5.2 scored within a point of Claude Opus 4.7 and 4.8 at about five times lower cost per completed task. Eight of OpenRouter's top ten models by August 2026 token volume ship open weights.[1]

The same 4.4-month gap is the second write-up's lead

AI Chat Daily, citing Mozilla's report published 15 September, says the best Chinese open-weight models trail US closed frontier systems by 4.4 months. Moonshot AI's Kimi K3 trails Fable 5 by three points on the Artificial Analysis Intelligence Index at 30 percent of the cost. CTO Raffi Krikorian frames closed models as useful for expert work, high-intensity retrieval and long-context jobs, with the unique closed-model window at eight to 12 hours of expert human time. On a neutral Terminal-Bench 2.1 harness, GLM 5.2 landed within one point of Claude Opus 4.7 and 4.8 at about five times lower per-task cost.[2]

Both texts rest on the same Mozilla scoreboard

The 4.4-month METR horizon, Kimi K3's three-point gap to Fable 5 and GLM 5.2's Terminal-Bench near-tie sit in both write-ups. Ars Technica counts the OpenRouter open-weight share and the price band; AI Chat Daily foregrounds Krikorian's eight-to-12-hour expert window. Those are the report's own measurements; who leads in production is a separate question.[1], [2]

References

  1. News sourceArs TechnicaMozilla: the open-weight gap with closed frontier models is 4.4 months↩1↩2
  2. News sourceAI Chat DailyMozilla puts the open-weight gap with closed frontier models at 4.4 months↩1↩2
AI safety and regulationChip wars and the supply chainThe frontier model race

More analysis from this desk

View all Brief analyses from the Artificial Intelligence desk →