Same base model, 6x agentic jump: GLM-5.3 hits 28.3% on Terminal-Bench via post-training
bittingthembits · x · 2026-08-25
GLM-5.3 keeps the same base model as GLM-5.2, but scaled post-training — more environments, more diverse tasks, more compute — lifting Terminal-Bench 3.0 resolution from 4.6% to 28.3%, roughly a 6x jump. Terminal-Bench tests agents on complex tasks in real terminal environments.
The poster argues pretraining (learning language, code, patterns) is dominated by big labs with massive GPU budgets, while post-training (tasks, environments, feedback, rewards) is a far more open race — which Affine, a Bittensor subnet, has turned into an ongoing open competition.
More from Models
- GLM-5.3 unprompted reasons about how its actions affect its future training, worries Apollo — MariusHobbhahn · 2026-10-15
- Gemini 4 argon vs Opus 5.5 hands-on: 3 min for $3 vs 25 min for $8 — zacharynado · 2026-10-15
- SentinelOne says OpenAI did 4-6 weeks of reverse-engineering in 8 hours — OpenAI · 2026-10-15
- Qwen3.8 27B scores 23.57% on word-based addition without reasoning, 167/169 with it — dl_weekly · 2026-10-15
- Hacker turns EmbeddingGemma 2 into 288x realtime STT and a fast TTS — kastnerkyle · 2026-10-15
- Claude Fable 5.5 reportedly in partners' hands, release eyed next week — lyraxana · 2026-10-15