#评测/基准

MiniMax m3 is a huge 26% improvement on BU Bench with browsercode, and shows promise for some potential future improvement. Now it is on the level of Claude 4.6-sonnet and gemini 3.5 flash

Minimax M3 is excellent at SVG generation, reaching close to Gemini 3.5 Flash levels and beating Opus 4.7 on SVG-Bench. With 1M context, native multimodality, strong agentic/coding ability and open weights coming soon, the closed-source moat is thinning fast. Full Video: https://youtu.be/p6Npi-HBoRU

Blown away by MiniMax M3 I’ve been using it for free inside @orca_build with the @opencode agent Mostly UI tasks & code review so far, but so far it feels on par with Opus-4.7 Didn’t expect it to be this good. (also free while it lasts)

Liquid's LFM2.5-8B-A1B smashed OpenAI's gpt-oss-20b on tool calling We ran both locally on a MacBook Pro M5 Max, 64GB, and gave each the same trip-planning request that only completes if the model fires all 7 tool calls - weather for 3 cities, two currency conversions, an email and a reminder Outputs: LFM2.5-8B-A1B: 4.8 GB RAM usage, 7/7 tool-calls, 266 tok/s, 6.9s OpenAI gpt-oss-20b: 11 GB RAM usage, 3/7 tool-calls, 146 tok/s, 15.0s The 8B used less than half the RAM and still fired all 7 calls, while the 20B silently dropped more than half of its own. It also ran ~2x faster, wrapping the full agentic request in 6.9s against 15s. That's what 38T training tokens buy: a 1B-active MoE that nails the agentic tool calls a model 2.5x its active size keeps dropping

I gave the same Go MMRPG backend spec to two coding agents and then asked Claude to judge the performance. CC (with Sonnet 4.6) 4.6 shipped 1,409 lines of clean, readable code. CyOps with @MiniMax_AI M2.7 shipped 5,271 lines of denser, layered code.

Thrilled to welcome Step 3.7 Flash landing on ModelScope, a 198B sparse MoE VLM from @StepFun_ai 🔥🤖 https://modelscope.cn/collections/stepfun-ai/Step-37-Flash 198B total, 11B active per token, up to 400 tok/s. 256K context. Three reasoning levels (low/medium/high) to balance speed, cost, and depth. 🥇 #1 on ClawEval-1.1 (67.1), leading Kimi K2.6 by 9.3pts 🥈 #2 on SWE-bench Pro (56.3) 🥇 #1 on SimpleVQA Search (79.2) 🖼️ Multimodal: 196B language backbone + 1.8B vision encoder. Parses dense UI, charts, and financial reports natively. 📄 Apache 2.0. vLLM / SGLang / llama.cpp / Transformers supported. Hermes Agent / OpenClaw / Kilo Code compatible. Runs locally on Mac Studio or DGX Station with 128GB+ memory.

http://x.com/i/article/2053566242338795520

Artificial Analysis and IBM Research are launching ITBench-AA, the first in a new series of benchmarks evaluating models on agentic enterprise IT tasks, starting with Site Reliability Engineering tasks where frontier models score below 50% ITBench-AA’s SRE tasks benchmark model performance on Kubernetes incident response, where models must diagnose live systems by reading logs, tracing dependencies, and identifying root-cause entities across complex infrastructure. The underlying ITBench dataset has been developed by @IBM's Software Innovation Lab, leveraging IBM’s deep expertise in enterprise IT operations Artificial Analysis has worked closely with IBM over the last 6 months to develop a implementation of the dataset for frontier AI evaluation, beginning with Site Reliability Engineering (SRE) and expanding to Financial Operations (FinOps) and Chief Information Security Officer (CISO) tasks over time ITBench-AA SRE overview: ➤ 59 SRE tasks in total: 40 public tasks and 19 brand new, held-out tasks ➤ Each task provides a Kubernetes incident snapshot containing alerts, events, traces, metrics, logs, and application topology. The model must identify the minimal set of independent root-cause Kubernetes entities responsible for the incident ➤ Faults span typical SRE failure modes including infrastructure, service, application, and chaos-injected incidents, such as resource quota exhaustion, rollout failures, connection pool exhaustion, and network partitions Methodology details: ➤ Agentic harness: each task is solved by the model running in our open-source Stirrup reference harness, with shell access to a sandboxed file system containing the relevant logs and snapshots. 100-turn cap per task, 3 repeats per task ➤ Models submit a list of root-cause entities (Kubernetes Deployments, Services, Pods, etc.) they believe caused the incident. Each submission is compared against a ground-truth set of root causes provided by IBM Research ➤ Scoring uses average precision at full recall: if a model misses any of the ground-truth root causes, it scores 0.0 for that repeat. If it identifies all of them, it is awarded a score equal to its precision - the share of its submitted entities that are actual root causes, i.e. true positives / (true positives + false positives). The headline score is the average across 59 tasks × 3 repeats. ➤ The harness (Stirrup) is held constant across all evaluated models, allowing an apples-to-apples comparison between models. Key findings: ➤ Claude Opus 4.7 (Adaptive Reasoning, Max Effort) leads at 47%, followed by GPT-5.5 (xhigh) at 46% and Qwen3.7 Max at 42% ➤ All frontier models score below 50%, making ITBench-AA SRE one of the least saturated agentic benchmarks in our suite. For context, frontier models score considerably higher on Terminal-Bench ➤ Turn counts vary nearly 3x and longer trajectories do not translate to higher accuracy. GPT-5.5 (xhigh) averages 31 turns per task at 46%, while Gemini 3.1 Pro Preview averages 83 turns at 30%. Models that over-investigate tend to surface upstream fault-injection mechanisms or co-occurring symptoms as false positives ➤ GLM-5.1 (Reasoning) leads open weights models at 40%, effectively tied with Gemini 3.5 Flash (high). DeepSeek V4 Pro (Reasoning, Max Effort) follows at 38%, with Gemma 4 31B (Reasoning) at 37%, ahead of Gemini 3.1 Pro Preview at 30%

Many developers have suspected for months that GPT-5.5 outperforms Claude Sonnet for coding. But SWE-Bench reported near-parity, and it made people question what they’d been seeing in practice. DeepSWE aligns more closely with that day-to-day experience: GPT-5.5 scores 70% versus Claude Sonnet at 32%. That difference is substantial. DeepSWE focuses on what tends to matter in real workflows: whether an agent can take a short behavioral prompt, locate the correct area of the codebase, and implement the change cleanly - without needing you to enumerate files, modules, and functions. SWE-Bench often fails to capture that, due to dataset contamination and weaker verification. https://deepswe.datacurve.ai/blog

Today we’re releasing DeepSWE, a new standard for agentic coding benchmarks. On public leaderboards, top models often look relatively close in capability. DeepSWE shows where they actually diverge, reflecting the realistic experience of developers in their day-to-day work.

Qwen3.7 Max (20250517) 在 Code Arena: Frontend 中首次亮相便位列第 4——是榜单上排名最高的中国实验室，超越了 GLM-5.1，如今在智能体网页开发任务上已与 Claude Opus 4.6 不相上下。 热烈祝贺 @Alibaba_Qwen 取得这一成就！