Topic: "agentic-reinforcement-learning"

Jul 13
not much happened today
Show details
gpt-5.6-sol grok-4.5 terra-max fable-5-max opus-4.8 100b-reasoning-model prime-intellect vllm langchain threepointone factory cognition arena artificial-analysis parlance-labs agentic-reinforcement-learning rollout-traces message-dags long-horizon-reinforcement-learning multimodality harness-design cost-per-task coding-agents benchmarks model-efficiency real-world-evaluation task-specialization johannes_hage willccbb mikasenghaas xeophon omarsar0 skirano imjaredz
Prime Intellect released verifiers v1, a redesigned environment stack for agentic reinforcement learning and evaluations, improving efficiency by storing rollout traces as message DAGs to reduce complexity from O(n²) to O(n). This enables practical long-horizon multimodal rollouts, demonstrated with a 100B reasoning model running 40-turn SWE agent tasks on 6 H200 nodes in under 2 days. The ecosystem support includes vLLM integration to avoid tokenization drift. Discussions highlight that harnesses are becoming critical as the product surface for coding agents, with task-specialized harnesses favored over generic wrappers. Benchmarks are shifting focus from token price to cost per task, with models like Terra Max, Fable 5 Max, and Opus 4.8 compared on efficiency and cost. Real-world agent benchmarks show GPT-5.6 Sol ranking #2 and Grok-4.5 jumping to #13 on Arena's leaderboard, emphasizing cost per task as a key metric for long-horizon knowledge work.
Dec 19, 2025
not much happened today
Show details
qwen-image-layered kling-2.6 gwm-1 gen-4.5 gemini-3-flash gpt-5.2 codex-cli opus-4.5 alibaba kling-ai runway google anthropic openai image-decomposition motion-control video-generation agentic-reinforcement-learning long-context model-degradation benchmarking tool-use prompt-engineering ankesh_anand
Alibaba released Qwen-Image-Layered, an open-source model enabling Photoshop-grade layered image decomposition with recursive infinite layers and prompt-controlled structure. Kling 2.6 introduced advanced motion control for image-to-video workflows, supported by a creator contest and prompt recipes. Runway unveiled the GWM-1 family with frame-by-frame video generation and Gen-4.5 updates adding audio and multi-shot editing. In LLM platforms, Gemini 3 Flash leads benchmarks over GPT-5.2, attributed to agentic reinforcement learning improvements post-distillation. Users note GPT-5.2 excels at long-context tasks (~256k tokens) but face UX limitations pushing some to use Codex CLI. Discussions around Anthropic Opus 4.5 suggest perceived model degradation linked to user expectations.