devotd.
Recent developments, checked against their sources.
QwenGyre moves GPU capacity between agent rollouts and reinforcement-learning updates as work finishes. In a 48-step NL2RepoBench run with a 2.4T model, it finished 59.13 hours (43.9%) before Async and 16.05 hours (17.5%) before Colocate. The much-quoted 1.85× result belongs to a different 122B TerminalBench test; the paper provides no code link or repeated-run variability.
Parsimony counts how much code the patches from public coding-agent benchmarks add. The website ranks by mean net coding units over measurable solved attempts, with the solve rate and measurement coverage shown in separate columns.
TRL v1.15.0 scores tokens in SFT, DPO, KTO, GRPO, RLOO and Distillation without building the full logits tensor. In the project's measurements DPO's maximum trainable length grew 5.80×, but the effect depends on the previous loss path and vocabulary size.
On October 5 Reflection announced Beam, a 501B-parameter MoE model, and began opening its beta API gradually. For moving an existing agent, what matters is the service token limits, the supported endpoints and parameters that are accepted but have no effect.
Ai2’s AstaBrief 8B is a Qwen3-8B-based model that turns a question and retrieved literature excerpts into a cited report in one pass. Evidence retrieval belongs to the input path, and while the model is Apache 2.0, the prompt and SFT/DPO data are CC BY-NC 4.0.
On October 1 Ai2 released Olmo-core 3, a framework with a redesigned mixture-of-experts (MoE) training system. Its trillion-scale tests show the system performance of the training stack, not the quality of a newly trained model.
Claude Sonnet 5.5 keeps Sonnet 5’s base input and output rates, but changing the model ID alone can break a Messages API integration. Disabled thinking and forced tool_choice any/tool return HTTP 400; applications must also handle thinking blocks rather than assume the first response block is text.