Open weights models, via GLM 5.2, had their "very practically useful" in coding harness moment before Gemini. ~200 days since the release of Opus 4.5.
7:00 AM · Jun 21, 2026 · 10.1K Views
AI2 post-training lead Nathan Lambert says open-weights models achieved practical coding utility ahead of Google's Gemini
Story Overview
Nathan Lambert, AI2's post-training lead, highlighted that open-weights models like GLM 5.2 have reached a stage of real coding and agentic usefulness ahead of Google's Gemini, with the milestone landing roughly 200 days after Anthropic's Claude Opus 4.5 launch.
111566129.2K
Developer Impact
Benchmarks show open models pulling ahead on specific tasks
GLM 5.2, a 744B-parameter MoE model from Z.ai with a 1M-token context window, posted leading open-weights scores on coding arenas and agentic benchmarks, though direct head-to-head data against Gemini remains limited to those public leaderboards.
Open Question
Questions linger on capability versus ecosystem reach
Observers note it is unclear whether Gemini trails in raw ability or simply faces narrower adoption for coding workflows, with no further details available on timelines or broader performance gaps.
Sentiment
Users are reacting to open weights models reaching practical coding usefulness before Gemini, with some praising their rapid progress while others call Gemini mediocre and express sadness over Google's lagging position.
Pos
28.6%
Neg
71.4%
7 comments with sentiment.
Cluster Engagement
Digg Deeper
No Digg Deeper questions have been answered for this story yet.
Posts from X
Most Activity
Most Activity
VIEWS245
xlr8harder@xlr8harder
@natolambert damn, that's a good point.
though I don't know if gemini really isn't there or if they have just managed to gather zero interest in adoption of their ecosystem.
Nathan Lambert@natolambert
Open weights models, via GLM 5.2, had their "very practically useful" in coding harness moment before Gemini. ~200 days since the release of Opus 4.5.
1hViews 245Likes 2Bookmarks 0
LIKES5
Nathan Lambert@natolambert
@liyucheng_2 sir be realistic, its a community consensus
2hViews 230Likes 5
REPLIES2
Yucheng Li@liyucheng_2
@natolambert wrong, that's glm5.1, 5.2 is another leap
2hViews 219Likes 1
Adam Goldstein@goldstein_aa
@natolambert I am a Google shareholder. This makes me very sad. Can Google recover, or is it too late already?
2hViews 199Likes 1
Kelvin09@Onwuta_Kelvin
@natolambert What's the least spec of GPU I need to run GLM 5.2 locally
1hViews 38
Quentin Compsci@QuentinCompsci
@Onwuta_Kelvin @natolambert 4x 128GB Nvidia Spark for not usable token per second. ~$12k
1hViews 1Likes 1
Wilkins Micawber@Me5466255992308
@natolambert What is your own experience with GLM 5.2 in a coding harness?
1hViews 153
Da7em@Da7_Tech
@natolambert
1hViews 146
Sahebjot Singh@zereraz
@natolambert kimi 2.5 was a moment as well, multi modal and fast.
1hViews 74
Evan Ross Davis 🇺🇲@evanrossdavis
@natolambert Gemini doesn't seem like a serious model at all compared to GPT-5.5. Feels like some mediocre open model to be frank.
1hViews 22
安叫兽|Bird🕊️ 🔶 BNB@ajs6888
@natolambert 开源模型这波追得真快，编码场景最明显
1hViews 21
loanyen scipio nissim@scipio_nissim
@goldstein_aa @natolambert I’m with you. Feel your pain. But Google has deep pockets and a lot of talent.
Let’s see how the next iteration of Gemini performs. My investments also hope it can get back onto the frontier.
1hViews 15
Kklish@Kklishseru
@natolambert @liyucheng_2 4.5已经很好了，考虑到glm5.2大小只有753B，我相信中国很快就可以做出fable级别的，阿迪王那个傻逼用fable恐吓全世界简直太反人类了
1hViews 12
V0LYX@0xV0LYX
@natolambert its been ~200 days of everyone pretending open weights was supposed to keep pace
the utility gap is just longer than people wanna admit
35mViews 10
Epicarism@epicarism
@natolambert Insane
15mViews 6
Mike Frager ⧉@fragermk
@liyucheng_2 @natolambert The kabobs don’t lie.
1hViews 6
Kelvin09@Onwuta_Kelvin
@QuentinCompsci @natolambert Bruh!!
1hViews 1