#post-training
Post-Training: 9 AI articles covering post-training news, analysis, and research
Articles
Plurai Wants to Make AI Agents More Reliable Using Self-Drivingâ9
Plurai raises $13M to boost AI agent reliability using self-driving car simulation techniques, testing models with synthetic data to handle edge cases.
Trajectory Raises $45M to Make AI Model Training Accessible toâ9
Trajectory raises $45M Series A at a $300M valuation to make AI model training accessible to nontechnical experts through continual learning, helping models imp...
Improving OCR Faithfulness via Gated and Attenuated On-Policyâ8
GAD-RL adaptively gates and attenuates teacher supervision during on-policy distillation, improving OCR faithfulness and achieving strong CHAOS-Bench and OmniDo...
Fireworks AI Releases Ember-1: A Post-Trained Kimi K3 That Usesâ8
Fireworks AI's Ember-1 post-trains Kimi K3 to cut reasoning tokens by ~40% while preserving accuracy, available only via serverless API research preview.
Perplexity Trains Its Computer Agent on Real Mistakes With Hintâ9
Perplexity trains its computer agent on real mistakes using hint-guided self-distillation, cutting tool-call failures 21.2% in live testing.
Cognition Releases SWE-2: A Kimi K3 Post-Trained Coding Modelâ7
Cognition's SWE-2 coding model, post-trained from Kimi K3, hits 50.0% on FrontierCode 1.1, matching Fable 5.1 at 64% lower cost and running only inside Devin.
Z.ai Ships GLM-5.3 Without Retraining the Base Model: Better atâ8
Z.ai's GLM-5.3 boosts coding and cybersecurity via post-training, no retraining. Scores surge on complex benchmarks; open weights release in two weeks.
SpaceXAI Releases Grok 4.6: A 500K-Context Frontier Model Tunedâ7
SpaceXAI unveils Grok 4.6 with 500K context, boosting agentic coding and knowledge work with improved scores and new xhigh reasoning tier.
AllenAI Open Instruct Tulu 3: Building a Compact Post-Trainingâ9
Learn to build a compact post-training pipeline with AllenAI's Tulu 3: SFT, DPO, and RLVR using GRPO in a single-GPU environment.