#Model release

AIIndexTeam has released Index-Nailong-2B-FP4, an official NVFP4 (W4A4) quantization of its Index-Nailong-2B multilingual translation model, which supports 150 languages. The checkpoint keeps lm_head, embeddings, and MoE router gates in BF16, and a perplexity test on a fixed corpus rose from 3.2806 to 3.4998 (+6.68%), while zh->en and en->zh outputs matched BF16 semantically. Full FP4 acceleration requires an NVIDIA Blackwell GPU; on Hopper or Ampere, vLLM provides only memory savings, so the FP8 build is recommended.

AIIndexTeam released Index-Homura-9B-FP4, an official NVFP4 (W4A4) quantization of the Index-Homura-9B translation model from the Index-Translate family. On a fixed corpus, perplexity rose from 2.5386 in BF16 to 2.6245, a 3.38% increase, and zh->en generations matched the original. Full FP4 compute acceleration requires an NVIDIA Blackwell GPU, while older GPUs get only weight-only memory savings and the FP8 build is recommended for them.

AIIndexTeam released Index-Homura-2B-FP4, an official NVFP4 (W4A4) quantization of its Index-Homura-2B multilingual translation model, which supports 150 languages. The quantized checkpoint shows a 5.73% perplexity increase over the BF16 original (3.5011 to 3.7017) on a fixed corpus, and its zh-en and en-zh outputs are semantically equivalent under greedy decoding. Full FP4 acceleration requires an NVIDIA Blackwell GPU, while the source recommends the FP8 build for Hopper and Ampere hardware.

AIAs of October 2, 2026, GitHub deprecated Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, and Claude Opus 4.7 across all Copilot experiences. Suggested replacements are Gemini 3.8 Flash, Kimi K3, and Claude Opus 5.5, and Enterprise administrators may need to enable them through model policies.

AIAi2 released AstaBrief 8B, an open-weights model that turns a research question and retrieved literature excerpts into a cited report, along with its training data. The model runs as Fast mode in Asta, averaging 51.1 seconds per report versus 178.5 seconds for Thinking mode, about 3.5x faster. The post also describes filtering synthetic training data by citation density and building DPO pairs judged by two models that agreed.

Why it matters: The post explains how supervised fine-tuning, preference data, and citation-density filtering were used to build a cited-report model, which is useful for teams training their own models.

AIAi2 released AstaBrief 8B, a model that turns a research question and retrieved literature excerpts into a cited report, along with its training data. In Asta's Generate a report feature, Fast mode averages 51.1 seconds per report versus 178.5 seconds for Thinking mode, about 3.5x faster. The model is built on Qwen3-8B with supervised fine-tuning and DPO, and institutions can run its open weights on their own infrastructure.

Why it matters: The post explains the data filtering and one-pass generation choices behind a fast open-weights report model, showing what worked and what did not.

AINVIDIA has released PixelUMM, an encoder-free unified multimodal model with 15,199,672,064 parameters that handles text, image, and video understanding and generation directly in pixel space. It represents images as 16-by-16 RGB pixel patches on a Qwen3-8B language backbone, with iterative denoising for generation. The checkpoint is licensed for non-commercial research or evaluation only, while the source code is under Apache License 2.0.

AICloudflare released Clef and Clef-flash, two decision models hosted on Workers AI and open-sourced on Hugging Face under Apache 2.0, and launched a reinforcement learning fine-tuning service. In Cloudflare's tests, Clef classified a domain in 2.2s versus 4.7s for gpt-oss-120b, and the models are Jev-API compatible. The company is offering fine-tuning first through a forward-deployed engineering team, with a self-serve platform planned later.

AIA flu forecasting model built with Google AI ranked first among 39 eligible models in the CDC's FluSight 2025-26 season evaluation for predicting U.S. flu-related hospital admissions. The model was developed using Empirical Research Assistance (ERA), an AI tool that generates optimization algorithms, and ERA's underlying technology is now available to trusted testers.

AIGoogle DeepMind announced Gemini 4 Argon, rolling out first to trusted cyber defenders through its Fairwind Program. Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with output limits raised to 1M tokens. The post cites a 77.9% score on DeepSWE v1.1 and 91.7% on LVBench, and says broad availability will follow safeguard testing.

Why it matters: The post pairs Argon's benchmark claims with the phased release, pricing, and safeguard details, helping readers weigh its frontier-level capabilities against its access limits.

AIGoogle announced Gemini 4 Argon, a new frontier model rolling out first to trusted cyber defenders through its Fairwind Program. The model's output limit rises to 1M tokens from 64K, and its introductory API price is $2 per million input tokens and $10 per million output tokens. Google says broader availability to developers, enterprises, and consumers will follow after more testing of guardrails.

Why it matters: The post pairs benchmark claims with a phased access plan, pricing, and safety measures, which helps readers judge how quickly Argon may reach developers.

AIKling 4.0 extends native single-pass video generation from 15 to 30 seconds and adds Multiple Keyframes supporting up to 10 keyframe images, versus Start & End Frames in Kling 3.0. It also expands reference inputs to up to 15 combined assets, including up to 5 videos totaling 30 seconds, and adds 10-bit HDR at 1080p and 4K. The all-new Kling 4.0 will officially launch in October, and Kling 4.0 Flash became available to a limited group of early-access users on September 28.

AIKling AI says Kling 4.0 will support native video generation from 3 to 30 seconds, with up to 10 keyframes for timing control. The full model is scheduled for October 2026, with early access to the new creation page rolling out from September 28.

AIKorean AI lab Upstage has released Solar Mini 4, a proprietary reasoning model that scores 24 on the Artificial Analysis Intelligence Index with 35B total and 3B active parameters. It is priced at $0.10/$0.40 per 1M input/output tokens and has a 1M-token context window, but averages 7.1 minutes per task due to heavy output token use. Its weights are not released, and its size cannot be independently verified.

AIArtificial Analysis reports that Google's Gemini 4 Argon scores 53 on its Intelligence Index with high reasoning, matching GPT-6 Astra (max) and one point ahead of GPT-6.1 Sol (max). At the current 50% launch discount, its cost per task is $1.99, about 60% of GPT-6 Astra's $3.26, but the discount's end date is unconfirmed and standard pricing would raise it to $3.98. The model is being rolled out to selected users and is not publicly available.

Why it matters: The benchmark compares Gemini 4 Argon's cost per task and hallucination rate with GPT-6 Astra, showing where its value depends on a temporary 50% discount.

AIKling 4.0 is entering early access, with a wider rollout planned for October, and Kling 4.0 Flash opens to Ultra Yearly subscribers on September 28. The update generates videos up to 30 seconds in a single pass, accepts up to 15 reference assets, and supports up to 10 keyframe images. Upcoming features include 10-bit HDR output at 4K and 1080p and video extension up to 2 minutes.

Why it matters: The post specifies concrete capability limits such as 30-second native generation, up to 15 references, and 10 keyframes, which help users judge fit for production workflows.

AIBAAI released AREX-2, a 27B-parameter long-horizon agent model that improves solutions over multiple test-time rounds by proposing, measuring, reflecting, and revising. It was trained on machine-learning and algorithmic-programming tasks with verifiable feedback, and the source reports that this self-improvement transfers to deep research. The model is Apache License 2.0 licensed and has a 262,144-token context length.

Why it matters: The source compares AREX-2 against closed and open models on coding and deep-research benchmarks, showing how test-time self-improvement is measured across task types.

AIInternLM's AutoVerifier, built on Qwen3_5MoeForConditionalGeneration with about 68 GiB of weights across 40 safetensors shards, evaluates natural-language mathematical proofs, explains errors, and identifies the earliest incorrect step. It serves as the automatic grader for AdvancedMathBench's ProverBench, which accepts a proof only when all eight judgments report -1. The model is a learned grader rather than a formal proof checker and can make errors.

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.

AIGoogle Cloud argues startups should combine open-weight models with frontier APIs rather than routing every request to one frontier model. It cites Gemma 4, which spans five sizes including a 31B dense model and a 26B A4B Mixture-of-Experts model, released under Apache 2.0. The article's examples report a 44% latency drop for Cue, from 876 ms to 488 ms, and a $0 server cost for BetterSpeak's on-device Gemma 4 E2B.

AIAnthropic has launched Claude Sonnet 5.5, the second model in its Claude 5.5 family. The company describes it as a faster, lower-cost complement to Claude Opus 5.5, and points readers to a blog post for more information.

Why it matters: The release note places Sonnet 5.5 beside Opus 5.5 in the Claude 5.5 family, clarifying which model suits speed and cost needs.

AIAmp now uses Claude Opus 5.5 for its medium mode by default, replacing GPT-5.6 Sol, while ChatGPT subscribers can keep medium pinned to GPT-5.6 Sol. In Amp's internal evals, Opus 5.5 solved 65% of tasks versus 61% for GPT-5.6 Sol and 56% for Opus 5, at lower cost, and it runs at high reasoning effort because xhigh and max cost more without scoring better.

Why it matters: The source reports internal eval scores, cost comparisons, and usage guidance for choosing reasoning effort, helping developers decide which model and setting to run.

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.

AIXiaomi has released MiMo-V2.6-Pro-MOPD, an upgrade of the MiMo-V2.6-Pro-RL checkpoint that fuses several domain-specialized teachers into one model via MOPD2 and targets tool-call repetition. The sparse MoE model has 1.02T total and 42B activated parameters, a 1M-token context length, and accepts text, image, video, and audio inputs. Weights are available on Hugging Face and ModelScope, with deployment recipes for SGLang and vLLM.

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.

AIAnthropic researchers used Claude Science to compute the nine-loop six-particle amplitude in planar N=4 super Yang-Mills, a toy-model result that physicist Lance Dixon checked. The work reportedly cost roughly one or two thousand dollars, with about $100 of compute for the bootstrap calculation, and a similar result was reached by Song He's group.

Why it matters: The guest post shows a frontier physics calculation done with modest compute, which helps readers gauge what current AI can handle in research and what it still cannot.

AIGoogle introduced Gemini 3.8 Live with Live Avatar, which adds a visual persona with lip-syncing and expressions to its live dialogue models. The feature is available in Gemini Enterprise and supports 97 languages, with custom avatars available through enterprise allowlisting. Google says all output is watermarked with SynthID.

Why it matters: The post specifies enterprise availability, custom avatar allowlisting, and 97-language support, which clarifies who can use the feature and how far it reaches.

AIGoogle says Gemini 3.8 Live with Live Avatar is now generally available in Gemini Enterprise, with US and EU endpoints, provisioned throughput, and enterprise compliance. Its video avatars use synchronized lip-syncing, custom avatars are limited to an allowlist, and generated audio and video carry SynthID watermarks. The model also understands and speaks 97 languages and can run tool calls in the background while the conversation continues.

AIinclusionAI has published public training-content summaries on Hugging Face for its Ling and Ring model versions, including Ling-2.0, Ling-2.5, Ling-2.6-1T, Ling-3.0, Ring-2.0, Ring-2.5-1T, and Ring-2.6-1T. The documents, organized under the template associated with Article 53(1)(d) of Regulation (EU) 2024/1689, contain documentation only, not model weights or training datasets. Each summary covers only the model versions it names.

AIGoogle Developers reproduced Ai2's Olmo 3 7B from scratch in MaxText on Google Cloud TPUs, covering both the stage-1 pre-training run and the stage-2 mid-training anneal. The match was checked on held-out C4 loss, an 8-task accuracy suite, multi-domain perplexity, and token-level KL, not just the training loss curve. The post also describes a data-loader bug that made training loss look better than the reference while held-out metrics did not move.

Why it matters: The post documents how a faithful reproduction was verified on held-out metrics, including a data bug that training loss alone would have hidden.

AIGoogle DeepMind introduced Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, text-to-speech models offering custom voice design, line-by-line performance control, and multilingual support across more than 100 languages. Flash TTS is rolling out to developers in the Gemini API and Google AI Studio and to everyone in Gemini Notebook, while Flash-Lite TTS is available to developers and in Google Vids. Voice replication requires consent verification, and generated audio carries SynthID watermarking.

Why it matters: The source details the voice design, performance direction, and consent safeguards, showing how the model covers creative and high-volume use cases with access across several Google products.

AIGoogle DeepMind's latest Gemini Audio models let developers design new vocal personas from natural language prompts, directing pacing, back channeling, and dialect shifts line by line. Developers can also recreate consistent adult voice profiles from a 30-second audio sample, with built-in consent verification, SynthID watermarking, and C2PA credentials.

AIBaseten has made NVIDIA Nemotron 3 Diarization available as batch, streaming, and real-time diarized transcription presets. The single checkpoint serves four algorithmic latencies from 0.32 to 30.4 seconds, and the post reports DER of 9.8% on AISHELL-4 at the low profile versus 27.2% for Streaming Sortformer v2.1.

Why it matters: The post shows one checkpoint serving four latency profiles with DER figures against named baselines, useful for judging real-time speaker labeling tradeoffs.

AIFireworks Research released Ember-1, a specialized model built on Kimi K3 that it says delivers the same quality with 40% fewer tokens. Across five industry benchmarks, Ember-1 matched K3 max quality at a fraction of the cost, and in two customer A/B tests it used about 35% fewer tokens per task. It is available as a Research Preview on Serverless, and Fireworks is also launching training support for customized models.

Why it matters: The source gives benchmark and A/B results for cutting reasoning tokens while holding quality, which bears on cost planning for coding and agent workloads.

AILovable now offers Opus 5.5, which it says matches Opus 5's results while finishing builds in a third to half fewer steps. Internal benchmarks showed Opus 5.5 scoring 4 to 6% ahead of Opus 5 on verification discipline, with step reductions of 26% to 57% and input token reductions of 21% to 59% across tasks.

AIBlack Forest Labs released FLUX 3 Action, an open-weights 7B world action model that outputs robot joint commands from camera frames, robot state, and a text instruction. On the RoboLab-120 benchmark it reports 42.92% task success, ahead of Cosmos3-Nano-Policy at 36.8% and π0.5 at 28.0%. The model is fine-tuned on DROID, is distributed under the FLUX Kommunity License v.1.0, and runs in about 32 GB of GPU memory in bfloat16.

Why it matters: The model card gives a benchmark comparison, parameter counts, and an action contract, so readers can judge how it compares with existing robot policies.

AIBlack Forest Labs has published FLUX 3 Action SO-101 on Hugging Face as an open-weights 7B world action model. It takes two camera frames, the robot state, and a text instruction, then returns the next 42 actions with predicted video frames, with 32 executed at 30 Hz before replanning. The card also provides a rank-32 LoRA fine-tuning recipe for user datasets and states that the application must enforce joint velocity, force, and workspace limits.