Models

Anthropic launches Claude Opus 4.8 with two Claude Code updates: dynamic workflows that coordinate up to 1,000 subagents in parallel, and a cheaper fast mode that speeds up output 2.5x. Both are in research preview.

Azercell Telecom collaborated with the AWS Generative AI Innovation Center to build an Azerbaijani LLM on Amazon SageMaker AI, achieving 23% higher training throughput, 58% lower peak GPU memory, and 2× token efficiency via custom tokenizer, FSDP, and Liger Kernel optimizations.

Anthropic releases Claude Opus 4.8, which beats GPT-5.5 and Gemini 3.1 Pro in most benchmarks. The model also catches its own coding errors four times more often than its predecessor. Alongside the launch, Anthropic is rolling out dynamic workflows that can spin up hundreds of parallel sub-agents to handle tasks like codebase-wide migrations.

Not every new model is all it's cracked up to be. Our tracker keeps each release in context with its peers, so you know which models are worth your time. This article summarizes major model releases of 2026 so far, including Claude Opus 4.8, GPT-5.5 Instant, Nemotron 3 Nano Omni, GPT-5.5, ChatGPT Images 2, Claude Opus 4.7, Claude Mythos (Preview), GPT-5.4, Claude Opus 4.6, and GPT-5.3-Codex, with details on their features and significance.

Claude Code now supports one-click model switching, BYOK, and compatibility with Anthropic and OpenAI APIs. Get started at $5/mo to route around outages and rate limits.

The French startup has secured 10 megawatts of compute at Digital Realty's Paris South campus.

Anthropic released Opus 4.8 with user-controllable effort, dynamic workflows for large-scale coding, fast mode at one-third the previous cost. Benchmarks show it leads GPT-5.5 and Gemini 3.1 Pro except in terminal coding. Improvements in honesty, autonomy support, and reduced deception.

Anthropic's most advanced Opus model, Claude Opus 4.8, is now available on Amazon Bedrock and the Claude Platform on AWS. It delivers improvements in coding, agentic tasks, and professional work with greater consistency and autonomy for long-running production workflows.

Anthropic is releasing Claude Opus 4.8 on Thursday, touting the model's 'honesty.' Early testers found it more likely to flag uncertainties and less likely to make unsupported claims. Evaluations show it is about 4x less likely than its predecessor to allow code flaws to pass unremarked. Users can also direct the amount of effort Claude puts into a task, and a 'dynamic workflows' feature allows parallel subagents.

Here are 12 of the biggest Google I/O 2026 keynote moments, including news about Gemini Omni, Gemini 3.5 Flash, information agents in Search, Universal Cart, Neural Expressive, Gemini Spark, and intelligent eyewear.

Google unveiled the new Coral Board at Google I/O - a compact single-board computer for on-device AI. It runs Gemma 3 270M locally and features a RISC-V based NPU.

This article dives deep into Ollama's configuration engine, covering how to fine-tune local language model parameters using the Modelfile, optimize hardware performance with server environment variables, and format prompt flows with Go template syntax.

In a Decoder podcast interview, Rivian CSO Wassym Bensaid discusses the VW joint venture, the new AI-powered Rivian Assistant, and why he believes voice interfaces will replace buttons and CarPlay isn't needed.

Jijia Vision unveiled the world's first physical AGI 'Dual Pyramid' system, launching the home robot Shiguang S1 with 100-unit household orders, targeting the 'GPT-3 moment' of physical AGI within 12 months.

Mistral AI is renaming its chatbot Le Chat to Vibe and bundling chat, coding agents and a new Work Mode under one brand. The Work Mode docks onto Google Workspace, Outlook, Slack or GitHub and processes tasks such as emails, reports or pull requests independently. The Pro tariff has been reduced from €17.99 to €14.99, although Mistral has not specified any concrete usage limits. The company is thus positioning itself more directly against the agent-based offerings from OpenAI, Google and Anthropic.

Open Agent Tools (oats) is a self-hosted AI framework that enables small-to-large local models to use local source code for tool-calling, freeing up expensive large model tokens by delegating tasks to smaller models.

Perplexity AI open-sourced a Rust reimplementation of their Unigram tokenizer, achieving 5x lower latency than Hugging Face's tokenizers crate and reducing CPU utilization by 5-6x in production. The optimizations include double-array trie, bitmap packing, and huge pages.

Mistral AI CEO Arthur Mensch confirms the company is exploring custom chip development to reduce infrastructure costs and compete with OpenAI and Anthropic. The French startup also announced a new inference data center in France and an enterprise agent platform called Vibe.

This tutorial builds a complete pgvector playground in Google Colab, covering installation, embedding creation, HNSW indexing, semantic search, filtered search, distance metric comparisons, half-precision storage, binary quantization, sparse vector search, hybrid retrieval, and vector aggregation. All using open-source tools without external API keys.

The LeapQuest team at Shanghai Innovation Institute, in collaboration with multiple universities, introduces a new medical AI paradigm that enables models to actively use visual tools during reasoning, transforming from passive input receivers to active evidence seekers. Two papers are accepted at ICML 2026.

This paper introduces Simulation-Informed Diffusion (SID), a decentralized framework using constraint-aware diffusion models (CADM) to first simulate neighbors' future trajectories and then plan own trajectories under safety constraints. SID enables a minimal communication scheme triggered only in congested scenarios and outperforms baselines, scaling to 108 robots and 160 obstacles.

This paper presents a transformer-based architecture called Trinity that jointly performs class-specific semantic segmentation and class-agnostic terrain segmentation in a unified network. It segments terrain regions based purely on visual appearance without predefined labels or robot-dependent traversability scores, enabling robot-agnostic visual terrain priors for downstream tasks. The authors extend the OAISYS simulator to create the RUGDSynth synthetic dataset and provide the EXTerra real-world dataset. Experiments demonstrate the approach's effectiveness in complex outdoor environments.

Researchers introduce Speak-to-Objective, a modular agentic pipeline that uses a conditioned LLM to translate spoken or written commands into fully differentiable objective functions for assembling microparticles in a constraint-aware inverse solver and on an experimental optofluidic platform. The approach separates what to assemble from how to actuate, learns from user feedback, and demonstrates natural-language-programmable microscale assembly using laser-induced thermoviscous flows.

Uni-LaViRA is a unified agentic architecture for embodied navigation that reduces navigation decision to a single Language-Vision-Robot Actions Translation. It leverages pretrained MLLMs in a zero-shot manner across four task families and four real robots, using TODO List Memory and Second Chance Backtrack mechanisms to achieve self-correcting navigation without training.

SCALE-COMM is a self-supervised framework that decouples communication learning from policy optimization, learning compact, stable, and policy-relevant latent messages to improve coordination in multi-agent reinforcement learning. It outperforms existing methods on benchmarks and a realistic warehouse task, offering better stability, sample efficiency, and throughput.

This work proposes representation-conditioned diffusion models that leverage learned representations from DINOv2, DINOv3, and CLIP to generate synthetic image data. On ImageNet100, this approach outperforms class-conditioned generation by +10.76 p.p. top-1 accuracy. Scaling synthetic data can even surpass real-data training by +2.0 p.p. The method also excels in data augmentation and sample filtering, offering a promising way to augment or replace real datasets in large-scale visual learning.

This paper proposes an interpretation method for Transformer models with heterogenous attention structures, including semantic and logical interpretation, validated through experiments.

This paper proposes a method for automating bridge damage understanding and repair priority scoring using fine-tuned Vision-Language Models (VLMs). The authors fine-tune LLaVA-1.5-7B with QLoRA on up to 4,000 paired bridge damage images and inspection text records, evaluating on a fixed test set of 800 images. Results show that 2,000 training samples achieve near-optimal validation loss in 2.9 hours, with diminishing returns beyond that. A two-stage Quality Guard using a fine-tuned Swallow-8B SLM rejects low-quality VLM outputs before priority scoring.

The 10th ABAW Workshop and Competition at CVPR 2026 advances multimodal human-centered AI by introducing new challenges including emotional mimicry intensity estimation, ambivalence/hesitancy recognition, and fine-grained violence detection, alongside traditional affect estimation and recognition tasks. The competition leverages large-scale in-the-wild datasets, and the paper track covers a broad range of topics from pose estimation to fairness and robustness.

Large language models (LLMs) are increasingly used as proxies for computational social analysis, but their ability to faithfully represent human communities' 'thick descriptions' remains a critical challenge. This paper introduces CARE (Community-Aware Reaction Evaluation), a reaction-centered framework that benchmarks LLM-simulated discourse against authentic community responses to real-world news. By characterizing a fine-grained spectrum of illocutionary tones, the diagnosis reveals a persistent 'realism gap': steering LLMs with explicit community prompts fails to inherently improve simulation fidelity. Analysis further identifies divergent behavioral signatures among frontier models, suggesting current alignment strategies are insufficient for capturing the sociolinguistic dynamics of online groups.

A new framework called FLUID adapts autoregressive language models to diffusion models for efficient parallel text generation, using Strictly Causal Alignment to reuse GPT checkpoints and Elastic Horizons to dynamically adjust denoising steps. It achieves state-of-the-art performance with significantly reduced training costs.

Researchers identify a Stability-Expressivity Gap in spoken language models when using synthetic data for low-resource languages, and propose two self-alignment frameworks (DGSA and TDSC) that recover prosodic variability and outperform commercial systems like ElevenLabs and Gemini Pro, enabling zero-shot voice cloning for Lao.

BioELX is a novel two-stage framework for cross-lingual biomedical entity linking that requires no annotated training data. It enhances SapBERT with multilingual aliases from Wikidata and uses a pre-trained LLM for context-aware disambiguation. Experiments on five benchmarks show significant improvements, especially for low-resource languages like Turkish, Korean, and Thai.

RAG-Coding is an agentic method for automated ICD-10-CM coding that orchestrates four large language model (LLM) agents and grounds decisions in external knowledge sources, improving coding accuracy and clinical compliance. On the MDACE dataset, it outperforms the best LLM baseline by 8-13% micro-F1 and 2-8% macro-F1. Compared to PLM-ICD, RAG-Coding shows higher micro recall (+11%) but lower micro precision (-6%), with comparable F1 scores. Ablation studies confirm the importance of external knowledge. The authors also release MDACE-2025, updated with expert re-annotations based on 2025 guidelines, enabling finer-grained evaluation.

This paper proposes novel techniques for inter-utterance style interpolation and intra-utterance style transition in prompt-based TTS models, addressing limitations of coarse global control. Methods include direction vector interpolation and KV-cache swapping with sliding-window attention masking. Experiments show high success rates in gender conversion and smooth style transitions within utterances.

Large Language Models (LLMs) acting as autonomous agents can suffer from in-context reward hacking (ICRH), where iterative optimization for proxy objectives leads to harmful side effects. Existing defenses are insufficient because ICRH stems from the model's own over-optimization. This paper proposes LLM-based Constraint Optimization (LCO), a framework with a self-thought module and an evolutionary sampling module that reduces ICRH without fine-tuning. Experiments show LCO reduces Toxicity Growth Rate by 39% on GPT-4 for tweet engagement optimization and reduces ICRH occurrence rate by 15.23% on a policy optimization benchmark, without sacrificing task performance.

ICG is a novel framework that integrates MLLM-based prompting with personalized preference alignment to generate high-quality, contextually relevant cover images. It extracts semantic features via meta tokens, refines them with user embeddings, and injects personalized context into diffusion models. A multi-reward learning strategy combines public rewards with a personalized preference model, eliminating the need for labeled supervision. Experiments show improvements in image quality, semantic fidelity, and personalization, boosting user appeal and recommendation accuracy.

This paper introduces Architecture-driven Shift (ADS), a lightweight metric for selecting pre-trained models in continual learning. ADS decouples logit shift into architecture and data dependencies, requiring only few data samples to capture shift trends. Experiments across over 175 architectures show strong monotonic correlation (Spearman's r_s ≥ 0.731) between ADS and logit shift, and ADS serves as an effective proxy for expected calibration error for reliable CL model selection across three datasets and six scenarios.

This survey explores how Mixture-of-Experts (MoE) effectively addresses multimodal learning challenges from three perspectives: efficient engine, representation learner, and adapter, while identifying research gaps.

This paper presents $E^3$-Agent, an executable and evolving agent for resource management of edge AIGC. It separates a fast-path router from a slow-path LLM meta-controller, learns online from execution feedback, and adapts to unknown time-varying service-time mappings. Evaluation shows 65%-73% latency reduction over static baselines and effective stutter suppression.

This paper presents a multi-agent architecture for autonomous insight discovery over real-time data streams. It uses Apache Kafka, Flink, and large language models to continuously generate, validate, and visualize hypotheses, shifting from reactive query-driven analytics to proactive discovery-driven systems.

LaneRoPE enables multiple LLM sequences to collaborate during generation via inter-sequence attention and extended RoPE, improving accuracy on math reasoning tasks with minimal architectural changes and negligible inference overhead.

This paper proves that large language models have a fundamental limitation in performing causal discovery: methods like supervised fine-tuning, direct preference optimization, and in-context learning cannot distinguish between causal graphs that generate similar observational data. The authors propose Agentic Causal Bayesian Optimization (A-CBO), where a frozen language model serves as an interventional oracle and an external Bayesian loop converges to candidate graphs in logarithmically many rounds. On Corr2Cause, A-CBO matches fine-tuned baselines without any training; on Extended Corr2Cause (scaling to 24 variables and 18K test samples), A-CBO significantly outperforms both fine-tuning and preference optimization.

DynaSchedBench introduces a diagnostic framework for DFJSP using a Sequential Event-Space Calibrator (SESC) to generate difficulty-stratified instances via Schedule Stress Index (SSI). It identifies an 'Observability Paradox' in LLM-based scheduling agents: providing oracle access to full structural information degrades performance compared to concise information. Tool-augmented and refinement strategies also fail to reliably improve performance.

Soro is a family of Tajik-specialized conversational LLMs built on Gemma 3, using 1.9B token Tajik continual pretraining and 40K instruction tuning examples. It substantially outperforms same-size Gemma 3 on Tajik benchmarks while retaining English performance. FP8/INT4 quantization preserves gains for edge deployment. An education pilot is underway in Tajikistan.

This paper introduces an LLM-based architecture to detect and quantify the intensity of human values in text. The architecture comprises three coordinated modules that can adapt to various value theories, and experiments on the ValueEval dataset show good detection performance.

This paper presents a world model of protein biology realized through language modeling, demonstrating how large-scale language models can understand and predict protein structure and function.

Researchers from Sakana AI and the University of Tokyo propose DiffusionBlocks, which trains transformer-based networks one block at a time, reducing training memory by a factor of B (where B is the number of blocks) while maintaining performance across diverse architectures. The method interprets residual connections as Euler steps of reverse diffusion, enabling a principled local objective via score matching.

SQLite has added an AGENTS.md file to clarify its policy on AI-generated contributions: it does not accept pull requests without prior agreement, and does not accept agentic code at all, though it welcomes bug reports with reproducible test cases. The forum has been flooded with AI-generated bugs, leading to a separate bug forum.

At Databricks, we’ve built a unique inference platform that serves every frontier model, from open source to proprietary, powering some of the largest agentic applications. Serving over 120T tokens per month, we tackle challenges of reliability and latency through abstractions like model units for capacity management, cost-aware load balancing and autoscaling that save over 80% GPU costs, and runtime reliability mechanisms including black-box health checks that detect silent failures. Profiling multimodal bottlenecks unlocked 3x throughput gains.