The open weights year
Twelve months that took open models from interesting to unavoidable, and where the gap actually sits.
#open-weights
7 pieces tagged open-weights,
from January 21, 2025 to December 19, 2025.
OpenAI ships open weights for the first time since GPT-2
gpt-oss-120b and gpt-oss-20b under Apache 2.0. A strategic reversal, six years late, and genuinely useful.
Kimi K2 is a trillion-parameter open weights release
Moonshot ships a 1T-parameter MoE with 32B active, tuned for agentic tool use, with weights you can download.
Qwen3 ships a whole family under Apache 2.0
Eight models from 0.6B to 235B, hybrid reasoning modes, and a genuinely permissive license across all of them.
Llama 4 arrives, and the leaderboard problem gets a name
Scout and Maverick ship with a 10M-token context claim and an arena entry that wasn't the released model.
The day the market priced in efficiency
Nvidia lost roughly $600 billion of market value in a session. The trigger was a paper about training costs.
DeepSeek R1 puts a reasoning model under an MIT license
Open weights, a published training recipe, and API pricing that reads like a typo. The reasoning-model moat just got a lot shallower.