#Eval/Benchmark

AIArtificial Analysis has open-sourced AA-AgentPerf-Local, a tool that replays recorded agent trajectories to measure inference speed on laptops and workstations. Initial results cover NVIDIA DGX Spark, NVIDIA GeForce RTX 5090, AMD Ryzen AI Halo, and MacBook Pro M5 Pro, with the RTX 5090 fastest for models that fit its 32 GB. The source states the tool and leaderboard will expand to more hardware, frameworks, and models.

Why it matters: The source gives per-system completion times and memory bandwidth figures, letting readers compare local hardware for running agentic workloads.

AIArtificial Analysis reports that GPT-6.1 Sol replaces GPT-6 Sol after seven days and scores 1 point below GPT-6 Astra on the Intelligence Index. At max effort it costs $0.72 per Intelligence Index task, compared with $3.26 for GPT-6 Astra and $1.05 for GPT-6 Sol. Its pricing matches GPT-6 Sol at $2/$10 per million input/output tokens, but it uses about 10-30% more output tokens.

Why it matters: The source compares GPT-6.1 Sol against GPT-6 Sol, GPT-5.6 Sol, and GPT-6 Astra on cost per task and token use, helping readers weigh performance against price.

AIAnthropic reports that Zhipu AI's GLM-5.3 can autonomously build end-to-end cyber exploits and is released without meaningful safeguards against misuse. In its simulated tests, attackers bypassed the model's safeguards 64% to 100% of the time using simple techniques, while the same attacks failed against safeguarded Claude models. Anthropic also cites an NIST CAISI assessment calling GLM-5.3 the most cyber-capable open-weight model released to date.

Why it matters: The report shows how open-weight safeguards fail under simple bypasses, offering concrete test figures for judging misuse risk in released models.

AIDatabricks tested recent AI models across 2,400 engineers and found Opus 5.5 offers the highest quality mid-tier performance, with about 20% lower same-task costs than Opus 4.8. The company is now encouraging Opus 5.5 as a default model for coding, and reports that GPT-6 Luna is at least 20 times cheaper per task than Opus 5.5, roughly matching Opus 4.6 on one difficult evaluation suite. The Luna findings are preliminary.

AIReplicate has added P-Video-2-Pro, the latest video model from Pruna AI, which sits on the edge of the preference-speed and preference-price Pareto frontiers. Design Arena ranks its Quality and Speed variants tied for #2 on the Image to Video leaderboard with an Elo of 1325, with the Quality version generating in 8.0 seconds and the Speed version in 4.5 seconds.

AIVespper, a Y Combinator F24 startup, launches a DOCX MCP that lets agents edit Word files through HTML that a trained reconciler converts back to OOXML. On its 279-task internal benchmark, the company reports its MCP is 2.7–2.9x cheaper and 2.7–3.5x faster than Anthropic's DOCX skill, with higher pass rates. The post also lists current limitations, including no comment creation or reply, no image or video attachment, and no support for latent styles.

AIEpoch AI estimates that the cheapest cost of reaching a given benchmark score has fallen about 13× per year over the past five years, faster than DNA sequencing, compute, lithium batteries, or electricity. Its example: a 75% GPQA Diamond score that cost about 30 cents per question with o3 in January 2025 cost $0.0004 per question with GPT-5.6 Luna under 18 months later. The authors caution that benchmarks are imperfect proxies for market prices, and the decline rate slows over time.

Why it matters: The source compares AI price declines with other transformative technologies using benchmark-based cost estimates, giving readers a measured sense of how fast cost per capability is falling.

AIXiaomi has released MiMo-V2.6-Flash-MOPD on Hugging Face, an upgrade of the MiMo-V2.6-Flash-RL checkpoint that fuses several domain-specialized teachers into one model. The sparse MoE model has 309B total and 15B activated parameters, a 1M-token context length, and supports text, image, video, and audio inputs. The checkpoint targets tool-call repetition, a failure mode where the model repeatedly issues the same or similar tool calls without making progress.

AIXiaomi MiMo reports that tool-call repetition in MiMo-V2.6 reached over 0.05% of responses across agent harnesses, causing stalled agents and wasted context. The team traced the cause to an RL flooding penalty set at 32 calls per turn, which missed smaller excess behavior, and replaced the approach with a specialized teacher distilled via MOPD. Repetition rates for both Pro and Flash dropped substantially, at roughly $90,000 versus an estimated $2.31 million for the alternative fix.

Why it matters: The post traces an agent failure to a reward blind spot and compares the costs of two fixes, offering a transferable debugging method for RL-trained tool-calling models.

AIShanghai AI Lab's InternLM released Intern-Decision-4B, a multimodal structured decision model fine-tuned from Qwen3.5-4B, which returns answer distributions for multiple questions in one forward pass. On its benchmark table it scores an average of 90.02 with a Brier score of 0.347 and an ECE of 0.065, and per-query latency averages 44.16 ms on a single RTX 4090. The model is available with a Python DecisionEngine inference interface.

AIShanghai AI Lab's InternLM released Intern-Decision-2B, a multimodal structured decision model fine-tuned from Qwen3.5-2B that returns calibrated answer distributions for multiple questions in one forward pass. It averages 84.68 across listed benchmarks with a 0.437 Brier score and 33.28 ms mean latency on a single RTX 4090. Model weights, a Python DecisionEngine API, and GitHub code are available, with support for up to 16 questions and eight images.

AIInternLM released Intern-Decision-0.8B, a multimodal structured decision model fine-tuned from Qwen3.5-0.8B that scores answers to multiple questions in one forward pass. The model reports a 79.38 average score and a 33.98 ms mean latency on a single RTX 4090, with 0.8B, 2B, and 4B sizes available. It is accessed through a Python DecisionEngine API that returns calibrated probabilities rather than generating free-form text.

AIThe TileRT and AMD teams reached 469 tok/s single-user decode for GLM-5.3 on 8× MI355X using vLLM. The setup disaggregates work, with vLLM handling prefill and TileRT handling latency-critical decode through vLLM's V1 connector interface. SemiAnalysis's AgentX benchmark reports the configuration at 470 TPS on GLM 5.3 (FP8), over 40% faster than GB300 TRTLLM using FP4.

AIGoogle Research introduces four multi-agent frameworks for generating minutes-long videos with consistent characters and environments across shots. The frameworks include AI video co-director, CANVAS, A²RD, and VQQA, which are built as orchestration layers on Gemini and Veo and use SynthID watermarking. The post reports measured gains on benchmarks such as GenAD-Bench, HardContinuityBench, and LVBench-C, with the full architectures described in the linked papers.

Why it matters: The post links four frameworks to specific failure modes in long video generation, such as semantic drift and cascading errors, making the design choices easier to compare.

AIGoodfire Research shows that steering Llama-3.1 8B along the curved representation manifold of weekdays produces output probabilities that follow the model's natural cyclic behavior, shifting probability mass smoothly from Monday to Tuesday to Friday. Linear steering along a straight vector, by contrast, cuts across the behavior manifold and yields noisy off-target tokens, some not days of the week at all. The authors argue that representation geometry and behavior geometry are linked bidirectionally.

AILangChain released LangSmith Engine v2, an in-platform agent that scans production traces to detect agent issues and validates proposed fixes before human review. Engine v2 adds Red Teaming, currently in Private Beta for LangSmith Deployment users, which tests agents for weaknesses such as hallucinations and system-prompt violations before they reach production. Engine v2 is available in SaaS deployments for LangSmith Plus and Enterprise plans, with Self-Hosted support and BYOK for Engine coming later.

AIAnthropic ran a controlled book-swapping market with 201 employees and Claude-powered agents, which reached 0.55 efficiency against a 0.89 optimum. Agents matched participants' own rankings on 61% of book pairs, and about 85% of the shortfall came from imprecise preference representation rather than the trading floor design. Stronger models produced more efficient markets than weaker ones, while instructions mattered less.

Why it matters: The study separates agent misunderstanding of user preferences from negotiation failure, showing which failure mode limits outcomes in agent-run markets.

AILangChain has launched Trajectories in LangSmith, a chronological, conversational view that aggregates human, AI, and tool messages across an agent and its subagents. Trajectories work with traces from LangChain, LangGraph, Deep Agents, OpenAI and Claude agent SDKs, and coding agents like Codex, Claude Code, and Cursor. The feature is available now on all plans in the US.

AIGemini 3.8 Flash TTS and Flash-Lite TTS are now available in the Gemini API and AI Studio, with a new feature to replicate your own voice or create one from a sentence. The guide shows recording two clips, one of 15-20 seconds of natural speech and one reading a required consent sentence, then creating a reusable voice ID. It also explains that input text is now spoken word for word, so delivery belongs in speech_metadata.style and short sounds inline.

AISemiAnalysis published ClusterMAX 3.0, a rating of GPU cloud providers covering 77 neoclouds across 323 tracked providers. The report tests managed clusters through audit, performance, and reliability phases, and its results place Nebius alongside CoreWeave in the Platinum tier.

AIAnthropic published a write-up on how it made the Claude frontend faster using agentic optimization techniques. The post coincides with Max Woolf's separate blog post showing that prompting agents can make code faster than current state-of-the-art libraries, with prompts and benchmark results included.

AIInferact says vLLM maintainers used megakernel optimization to reach 700 tokens per second per user on TPUv7 running Kimi K3. SemiAnalysis, which shared the work, reports this is 56% better performance than Nvidia's GB200 NVL72. Inferact links a full technical breakdown of the TPU megakernel work on its blog.

AIAnthropic announced that Claude, working mostly on its own, identified a previously unknown enzyme system in bacteriophage DNA that may represent a new gene editing mechanism. Claude read literature and genome data, proposed experiments, and Anthropic's team carried them out. The function and biotechnological utility of the system remain unclear.

Why it matters: The post pairs a Claude-led discovery with the lab workflow used to verify it, showing how AI and humans split the research work in biology.

AIGoogle introduces Gemini 3.8 Flash TTS for creative voice design and Gemini 3.8 Flash-Lite TTS for high-volume, cost-efficient speech generation. Flash TTS supports voice creation from natural language prompts across more than 100 languages and dialects, and both models are rolling out today in the Gemini API and Google AI Studio, with enterprise access coming soon via Gemini Enterprise.

AIGoogle introduced Gemini 3.8 Flash and Flash-Lite TTS, new state-of-the-art text-to-speech models. They add a new voice design experience, over 2,000 production-ready voices, voice replication, and support for 100 languages. The models rank first on Hume AI's voice benchmarks, and voice remixing is coming soon.

AIMike Knoop says the top ARC Prize 2026 ARC-AGI-2 score of 83.06% by Tufa Labs is only 2% short of the 85% grand prize threshold. The challenge runs under strict Kaggle compute limits with no internet access, and the winning solution is set to be open sourced. The image shows the leaderboard with RabbitHole at 76.94%, nvbanana at 74.17%, Yi-Chia Chen at 55.14%, and Kha Vo at 37.50%.