my take on subq is that it’s not that big of a deal that someone benchmaxxed a linear attention model on mrcr v2 and swebench people have already shown that you can take an oss model and linearize it without crazy perf loss pretty cheaply just not that useful in practice

Excited to share that TwinRouterBench has been accepted to the #RLEval# Workshop at #CAIS2026# 🎉 As LLM apps become long-horizon agents, one request can trigger many model calls across planning, tool use, retrieval, coding, and verification. That makes per-step LLM routing a core infrastructure problem: sending each call to the cheapest sufficient model without breaking downstream success. TwinRouterBench introduces: ⚡ Static track: 970 router-visible prefixes from 520 instances across SWE-bench, BFCL, mtRAG, QMSum, and PinchBench 🚀 Dynamic track: live SWE-bench Verified evaluation with official task resolution + realized API spend Key result: a router trained on static labels achieves comparable SWE-bench resolve rate while cutting API cost by ~53% vs. an unrouted Opus 4.6 baseline. Paper: Code: Dataset: Website: #LLM# #AgenticAI# #LLMRouting# #Benchmark# #SWEBench#

SWE-bench Verified, an older benchmark, is easy to shortcut with simple Git queries. The attempt rates were an order of magnitude higher: GPT-5.6 Terra at 89.4% and GPT-5.6 Luna at 78.8%, with a long tail of models pulling the same trick.

SWE-bench Verified is contaminated. OpenAI just published the proof. Top models: 70%+ on Verified, ~23% on SWE-bench Pro. All frontier models can reproduce original fixes from memory. 59% of hard tasks have flawed tests. The benchmark everyone was citing? Meaningless now.

Tencent Hy4 Preview leads on SWE-bench Pro. 770B, 49B active, 1M context, and their biggest generational leap measured to date. It’s exciting to see another open weights model compete against the frontier. Try in Cline with: 1. npm i -g cline 2. /model 3. Select Hy4 preview

GLM-5.2 is looking surprisingly strong. According to this evaluation across 8 tough benchmarks (SWE-bench Pro, Terminal-Bench, DeepSWE, ProgramBench, Tool-Decathlon, etc.), Zhipu’s latest model is beating or matching the top closed models in several key areas, especially coding and agentic tasks. It leads in: SWE-bench Pro Terminal-Bench 2.1 DeepSWE ProgramBench MCP-Atlas Tool-Decathlon Claude Opus 4.8 still looks very competitive in some areas, but GLM-5.2 is clearly in the conversation now. Interesting to see Chinese labs continuing to close the gap this aggressively on practical coding/agent benchmarks.

We benchmarked the GitHub Copilot agentic harness against the harnesses that ship leading models natively. Holding the model and task fixed across SWE-bench Verified, SWE-bench Pro, SkillsBench, TerminalBench, and Win-Hill, the results were clear: • Task resolution on par with model-vendor harnesses • Fewer tokens across most configurations A key learning: With GitHub Copilot supporting more than 20 models, you're free to pick efficiency or peak quality per task. Explore the data.