Google DeepMind Releases Genie 4, a Real-Time Interactive World Model
Google DeepMind has released Genie 4, a world model focused on real-time interactive simulation and stronger photorealistic realism; it also natively generates …
OpenAI Announces Three Scientific Breakthroughs by Its AI Models, Spanning the Local Smoothing Conjecture, the Exponential Time Hypothesis and Yang–Mills Theory
On October 27, OpenAI announced three research breakthroughs achieved by its AI models in mathematics, computer science and physics. In mathematics, the models …
OpenAI Releases GPT-6.1 Terra, Focused on Computer Use and Writing
The author announced the return of GPT-6.1 Terra, calling it their first model that is markedly better at writing, personality, and nuanced understanding of req…
OpenAI Releases GPT-6.1 Terra, Targeting Efficient Computer Use and Lower Costs
OpenAI has released GPT-6.1 Terra, a fast and efficient model for computer use and visual reasoning. Standard API pricing is 50% lower than GPT-5.6 Terra and GP…
Artificial Analysis benchmarks GPT-6.1 Terra: advancing the cost-performance frontier at under a third of the cost
Artificial Analysis evaluated OpenAI's GPT-6.1 Terra, saying it reaches the Pareto frontier of intelligence and cost at low-to-medium effort, priced at half or …
Kimi releases open-source flagship model K3.1 Pro: 2.8T parameters, 1M token context, tested on 3D and video motion effects
Moonshot AI has released its flagship model Kimi K3.1 Pro, featuring 2.8T parameters and support for a 1M token context, with full model weights expected to be …
Moonshot AI Releases Kimi K3.1 Pro With 1M-Token Context, Open Weights Due by November 4
Moonshot AI has released Kimi K3.1 Pro, the first model in the Kimi K3.1 series. It is a 3T-class open-weight model with a 1-million-token context window, and i…
MiniMax Opens Up M3.1 Flash Weights: 428B Total Parameters, 23B Activated per Token
MiniMax has announced that the M3.1 Flash weights are publicly available, with 428B total parameters and 23B activated per token, downloadable via Hugging Face …
MiniMax Open-Sources M3.1 Flash with 428B Total Parameters, Processing Over 100 Trillion Tokens Within 72 Hours of Launch
MiniMax has open-sourced the weights of the M3.1 Flash model, which has 428B total parameters and 23B activated parameters; within 72 hours of launch, it proces…
Kimi K3.1 Pro Released, Ranks 4th on Artificial Analysis Intelligence Index with 55 Points
Moonshot has released Kimi K3.1 Pro, which scores 55 on the Artificial Analysis Intelligence Index to take 4th place — an 11-point gain over Kimi K3, trailing o…
NVIDIA Expands Open-Source Model Family, Covering Language, Robotics, Biology and Quantum Computing to Support European Sovereign AI
NVIDIA is expanding its open-source model family, introducing Nemotron 3.5 Super VL and Nemotron 3.5 Bolt, Nemotron Retriever, the Kumo structured data model, t…
fal and bfl_ai Release Post-Training Video Model FLUX 3 Max
fal has released FLUX 3 Max, its latest post-training video model built in partnership with @bfl_ai. It tied for first place in a third-party image-to-video qua…
DeepSeek-V4.1-Flash-20261017 (Max) Takes the No. 1 Open-Source Spot on Agent Arena
Arena announced that DeepSeek-V4.1-Flash-20261017 (Max) has become the top open-source model on Agent Arena, ranking 11th overall, with a net gain of +5.1% base…
DeepSeek V4.1 Flash 1017 Released: Scores 45 on Artificial Analysis Intelligence Index, Tied for Second Among Open-Source Models
DeepSeek has released DeepSeek V4.1 Flash 1017, which scores 45 on the Artificial Analysis Intelligence Index — 6 points higher than DeepSeek V4.1 Flash. It tie…
Meta Muse Wonder Full (Max) Debuts on Code Arena: WebDev at No. 6 with 1,726 Points
Arena announced that Meta's Muse Wonder Full (Max) has entered the Code Arena: WebDev leaderboard for the first time, ranking 6th with 1,726 points, making Meta…
OpenAI Releases GPT-6 Series Models, Boosting Coding Capabilities in Low-Resource Programming Languages Like COBOL
OpenAI announced that its latest models show significant improvements in low-resource programming languages such as COBOL, Fortran, and Visual Basic, and introd…
Ant Group's inclusionAI Open-Sources Ming-3.0-tts Speech Synthesis Model
Ant Group's inclusionAI has released Ming-3.0-tts on GitHub, comprising two speech generation models — a 0.5b dense model and a tiny model built on Ling V3 (7.9…
Inception Labs Releases Diffusion LLM Mercury 3 at 1,068 tokens/s
Inception Labs has released Mercury 3, a diffusion-based large language model that it says outperforms GPT-6 Luna and Claude Haiku 5.5 on multiple benchmarks, r…
DeepSeek-V4.1-Flash-20261017 Tops Code Arena: WebDev Open-Source Model Rankings
Arena announced that DeepSeek-V4.1-Flash-20261017 (Max) has become the top open-source model on the Code Arena: WebDev leaderboard, surpassing Kimi K3 (Max) at …
MiniMax releases M3.1 Flash, delivering nearly 3x faster inference and 70% lower usage costs
MiniMax has released MiniMax M3.1 Flash. The anonymous model Space Bunny was an early version of it and has processed over 200 trillion tokens.
Muse Spark Open (Max) Tops Code Arena: WebDev as the No. 1 US Open-Source Model
Arena announced that Meta's Muse Spark Open (Max) ranks first among US open-source models on Code Arena: WebDev with a score of 1511, placing 13th overall on th…
DeepSeek-V4.1-Flash-1017 Released With Gains Across Multiple Benchmarks and Unchanged API Pricing
DeepSeek has upgraded V4.1 Flash to DeepSeek-V4.1-Flash-1017, with the web interface and app updated at the same time. The API calling method and pricing remain…
Kling 4.0 Goes Live: 30-Second Single-Take Support, Author Tests Motion and Omni-Reference
KlingAI officially launched its Kling 4.0 video model in the early hours of the morning, supporting direct output of up to 30 seconds and multi-frame reference …
Reflection Releases Open-Source MoE Model Beam 501B, Hugging Face Offers Deployment and Fine-Tuning Guide
Reflection's first open-source model, Beam (reflectionai/Beam-1.0-501B-A23B), is now live on Hugging Face. It is a 501B-parameter MoE model with 23B active para…
Muse Spark Open Lands on Arena, Now Testable in Agent Arena, Scores Coming Soon
Arena has announced that Meta's Muse Spark Open is now live in Agent Arena, where users can put it to the test; scores will be published shortly. The model feat…
Meta Releases Open-Weight Model Muse Spark Open with 816B Parameters and 1M Context
Meta has released the open-weight model Muse Spark Open, now available on Hugging Face under the Muse Model License. The model has 816B total parameters with 42…
Gemini 4 Argon Results Published on ARC-AGI (Verified): 97.5% on ARC-AGI-3
ARC Prize has published Gemini 4 Argon's results on ARC-AGI (Verified). On ARC-AGI-3, the provider adapter harness scored 97.5% at a cost of $4.8K, while the st…
Microsoft AI unveils MAI-Base-2 pretraining details: 2.8T-parameter MoE model reaches 4.2x to 12.3x the FLOPs efficiency of DeepSeek-V4-Pro-Base
Microsoft AI announced it has completed pretraining of MAI-Base-2, an 80B-activated, 2.8T-total-parameter MoE foundation model that achieves a lower BPB than ev…
Reflection Releases Beam, Its First Open-Weight Model, Scoring 30 on Artificial Analysis and Ranking Second Among US Open-Weight Models
Artificial Analysis evaluated Beam, Reflection's first model. It scored 30 on the Artificial Analysis Intelligence Index, ranking second among US open-weight mo…
Step 5 Preview Goes Live on FeatherlessAI; Official Account Shares Open-Weight Model Deployment Details
StepFun reposted FeatherlessAI's announcement that Step 5 Preview is now live on the platform, served at FP8 precision. The open-weight model uses a 600B sparse…
NVIDIA Releases Open-Source Nemotron 3.5 Super VL and Bolt On-Device Models
NVIDIA has released the open-source 120-billion-parameter Nemotron 3.5 Super VL multimodal model, which it says completes tasks faster than anything else in its…
Black Forest Labs Releases Open-Source Text-to-Image Model FLUX 3 Image Klein (5.6B)
Black Forest Labs has released FLUX 3 Image Klein, a 5.6B-parameter open-weight model that supports text-to-image, image editing, and multi-reference generation…
StepFun Showcases Step 5 Preview Inference Performance on NVIDIA GB300
StepFun says Step 5 Preview delivers excellent inference performance on NVIDIA GB300 and looks forward to seeing what developers build with it. The NVIDIA conte…
StepFun Opens Step 5 Preview Weights
StepFun has announced the open-weight release of Step 5 Preview. The BF16 and FP8 weights are now available on Hugging Face and ModelScope, along with an accomp…
StepFun Releases Step 5 Preview Weights and Publishes Technical Blog
StepFun has officially released the Step 5 Preview model weights, offering BF16 and FP8 versions, now available on Hugging Face and ModelScope.
Ant Ling Officially Open-Sources Ling-3.1-flash, a 560B-Parameter Model Deployable on 4 GPUs
Ant Ling has officially open-sourced Ling-3.1-flash, featuring roughly 560B total parameters, about 25B activated per token, and support for up to 1M context. T…
Kunlun Tech Releases and Open-Sources Matrix-Game 4.0, a Video Large-Model Game Running in Real Time on a Single RTX 5090
On October 15, 2026, Kunlun Tech released Matrix-Game 4.0, calling it the first video large-model game to run on a single consumer-grade GPU, and open-sourced i…
Gemini 4 Argon Opens to Google AI Pro and Ultra Subscribers Worldwide
Gemini officially announced that Gemini 4 Argon is now available to Google AI Pro and Ultra subscribers worldwide. The model had been rolling out gradually in t…
Musk Says Grok Voice 3.0 Tops Artificial Analysis Speech Model Benchmark
Musk posted that Grok Voice 3.0 is now the world's leading conversational voice model, citing Artificial Analysis's evaluation: Grok Voice 3.0 tops the Speech t…
Meta's Muse Wonder (Max) enters Agent Arena at No. 12 with a +6.4% net gain
Arena has released evaluation results based on over 5000 real agent sessions. Meta's Muse Wonder (Max) takes No. 12 in Agent Arena with a +6.4% net gain, scorin…
Artificial Analysis Benchmark: Grok Voice 3.0 Tops the Speech-to-Speech Model Rankings with 86.9 Points
Grok Voice 3.0 debuted at the top of the Artificial Analysis Speech to Speech Index with 86.9 points, ahead of GPT-Live-1 (Astra, medium) at 83.2 and Gemini 3.8…
Artificial Analysis Releases First AA-AgentPerf Agentic Inference Benchmarks for NVIDIA Vera Rubin NVL72
Artificial Analysis has published the first AA-AgentPerf results, testing the serving efficiency of NVIDIA Vera Rubin NVL72 under agentic coding workloads, comp…
Gemini 4 Argon is fully open. After actual testing, the author believes that it is difficult to shake Claude Opus 5.5 and GPT-6 Astra.
Google fully opened Gemini 4 Argon on October 14. After actual testing, the author believes that it is difficult to impact the status of Claude Opus 5.5 and GPT…
Meta Muse Wonder (Max) entered the Arena list: WebDev No. 15, Text No. 2
Arena announced that Meta's Muse Wonder (Max) entered the Code Arena: WebDev list at No. 15 (1651 points), 4 points behind Kimi K3 (Max)'s 1655 points.
Gemini 4 Argon is officially GA, the smart index is equal to GPT-6 Astra, and the Cost per Task is only 60%
Gemini 4 Argon officially entered GA, ranking the same as GPT-6 Astra on the Artificial Analysis Intelligence Index, with a Cost per Task of 60% of the latter.
Google releases Gemini 4 Argon and gradually pushes it to Gemini applications
Google Gemini announced that the new cutting-edge model Gemini 4 Argon is gradually rolling out in the Gemini application, positioning it to turn complex ideas …
Google DeepMind releases Gemini 4 Argon robot model, ERQA benchmark score 87%
Google DeepMind released Gemini 4 Argon, positioned as a general physical reasoning model, with a pure reasoning score of 87% on the Embodied Reasoning (ERQA) b…
Google releases Gemini 4 Argon: a cutting-edge model for complex engineering and scientific research tasks
Google releases Gemini 4 Argon, which is open to developers, enterprises, scientists and consumers globally and can be accessed through Gemini API, AI Studio, a…
ModelBestOpen source MiniCPM-V 4.7, 35B-A3B MoE model 3GB memory runs smoothly on both ends
ModelBest jointly released and open sourced MiniCPM-V 4.7 with the OpenBMB community on October 13, including two versions: 35B-A3B and 1B, calling it the world…
Liquid AI releases LFM2.5-24B-A2B device-side inference model
Liquid AI releases the open source weight model LFM2.5-24B-A2B, with a total parameter of 24B and activation of 2B per token, to position the end-side fast agen…
Meta Muse Wonder series launches on Agent Arena, scores will be announced soon
Arena announced that Meta's Muse Wonder series models are launched on Agent Arena, users can test them in practice, and the scores will be announced soon.
Meta releases Muse Wonder and Muse Wonder Full models, featuring personal and professional AI capabilities
Meta has released the Muse Wonder series of models, claiming that they achieve SOTA performance on personal and professional agent tasks. Muse Wonder is now ava…
Meta releases Muse Wonder and Muse Wonder Full; Artificial Analysis tests show intelligence level significantly improved compared to previous generation
Meta released two models, Muse Wonder and Muse Wonder Full, which scored 52 and 56 respectively on the Artificial Analysis Intelligence Index v4.3 (max reasonin…
Arc Prize 2026: TUFA Labs tops the ARC-AGI-2 highs with 88.06%
The arc Prize announces the ARC-AGI-2 high scores for the 2026 season, with TUFA Labs ranking first with 88.06%. $150,000 Bonus Prize in addition to $100,000 wi…
Mistral Large 4 entered the top fifteen labs in Agent Arena, ranking 43rd
Arena announced that Mistral Large 4 has entered the top fifteen labs of Agent Arena. Based on more than 5,000 real agent sessions, the preview version has a ne…
Arena announced the real evaluation results of Claude Haiku 5.5 (High): Code Arena scored 1587 points and ranked 30th in the debut
Arena announced the real evaluation results of Anthropic Claude Haiku 5.5 (High), which debuted at 30th place in Code Arena: WebDev with 1587 points, slightly o…
Artificial Analysis evaluates Google Nano Banana 2.1, ranking 4th on both lists and half the price of the previous generation
Artificial Analysis evaluates the image model Nano Banana 2.1 released by Google on October 6, ranking 4th in both AA-Image-T2I v2.0 and AA-Image-Editing v2.0 l…
Anthropic releases Claude Haiku 5.5, Artificial Analysis AI index score 43
Anthropic releases Claude Haiku 5.5, scoring 43 points on the Artificial Analysis intelligence index, an increase of 26 points in one year compared to the previ…
Perplexity open-source pplx-embed-v2-late multimodal late-interaction embedding model (9B and 0.6B)
Perplexity has open-sourced pplx-embed-v2-late, two late-interaction multi-vector embedding models for text and images, with sizes of 9B and 0.6B, sharing the s…
Mistral Large 4 entered Code Arena: WebDev, ranked 45th, with a score of 1534
Arena announced that Mistral Large 4 has landed on Code Arena: WebDev, ranking 45th with 1534 points, 304 points higher than Mistral Large 3 (ranked 130th) and …
Liquid AI releases open-source decision models d1-3B and d1-omni-600M
Liquid AI has released the open-source weight decision models d1-3B and d1-omni-600M, which are now available on Hugging Face.
Google DeepMind releases open-source lightweight multimodal embedding model EmbeddingGemma 2
Google DeepMind releases EmbeddingGemma 2, based on the Gemma 4 architecture, open-sourced under the Apache 2.0 license, mapping text, code, images, video, and …
Anthropic releases Claude Haiku 5.5, with average running costs reduced by about 75%
Anthropic released Claude Haiku 5.5, positioned as the cheapest and fastest small model to date, suitable for high-throughput tasks such as summarization, compr…
Anthropic releases Claude Haiku 5.5, Artificial Analysis review score 43
Anthropic released Claude Haiku 5.5, which scored 43 points on the Artificial Analysis Intelligence Index, an increase of 26 points within one year from the pre…
DeepSeek V4.1 Flash Performance Results Released in ARC-AGI (Verified) Evaluation
ARC Prize announced the results of DeepSeek V4.1 Flash on ARC-AGI (Verified): ARC-AGI-2 scored 72.9%, with a cost per task of $0.13; ARC-AGI-1 scored 94.5%, wit…
Mistral Large 4 Launches on Arena: Agent Arena and Code Arena Evaluation
Arena announced that Mistral's Mistral Large 4 is now available in Agent Arena, and users can go test and vote, with scores to be announced later. This model is…
Mistral releases the public preview of Large 4, with 1 trillion parameters, the weights will be open
Mistral has released the public preview of Mistral Large 4 (codename le Chonk), a native multimodal model with 1 trillion parameters and 49 billion activated pa…
Mistral releases Mistral Large 4: 1T parameters, natively multimodal, open weights by month's end
Mistral AI has released Mistral Large 4 (also known as Le Chonk): 1T parameters, 49B active, natively multimodal. The company says it is the best on aggregate b…
Gemini Nano Banana 2.1 officially released, gemini-3.1-flash-image will be deprecated on October 29, 2026
Google released Gemini Nano Banana 2.1 (gemini-nano-banana-2.1), positioned as an efficient image generation and conversational editing model, and it is an upda…
Mistral releases Mistral Large 4, Artificial Analysis review calls it the most intelligent model outside the US and China
Mistral released Mistral Large 4 (Research Public Preview), scoring 38 on the Artificial Analysis Intelligence Index, comparable to GPT-6 Luna (max, 38), making…
Liquid AI releases d1 decision model and adds image input capabilities
Liquid AI releases the d1 decision model with new text and image inputs, available through console.liquid.ai and d1 Playground.
LMSYS releases the open source chat model Vicuna-13B, which uses ShareGPT dialogue to fine-tune LLaMA. The training cost is about $300.
The LMSYS team released Vicuna-13B, fine-tuning LLaMA through conversations shared by about 70K ShareGPT users. The training cost is about $300, and the code, w…
GPT-6.1 Sol (Max) enters the 5th place in Agent Arena, approaching the top models at a lower cost
Arena announces that OpenAI’s GPT-6.1 Sol (Max) ranks 5th in Agent Arena (+11.23%) with a median task cost of $0.56 and reshapes the Pareto frontier.
Arena Review: Claude Sonnet 5.5 Tops Agent Arena at 3rd but Not on the Pareto Frontier
Arena released the latest list of Agent Arena. Anthropic's Claude Sonnet 5.5 ranked third with a +12.5% net improvement score. The median cost of a single tas…
Ai2 open source 8B scientific report generation model AstaBrief
Ai2 open source AstaBrief 8B, a scientific report generation model based on Qwen3-8B that converts research questions and retrieved literature fragments into ci…
Ai2 open source 8B scientific report generation model AstaBrief
Ai2 open source AstaBrief 8B, a scientific report generation model based on Qwen3-8B that converts research questions and retrieved literature fragments into ci…
Artificial Analysis: Claude Sonnet 5.5, GPT-6.1 Sol and Gemini 4 Argon top the Coding Agent Index list but the cost difference is large
Artificial Analysis released the Coding Agent Index list. Claude Sonnet 5.5 (max) ranked first with 68 points in Claude Code, but the cost per task was as high …
NVIDIA explains how Blackwell GPUs accelerate OpenAI GPT-6 Astra Ultrafast
GPT-6 Astra Ultrafast is now available in the OpenAI API and eligible ChatGPT Work and Codex users, running on NVIDIA Blackwell GPUs.
Artificial Analysis Review: Qwen-Image-2.1 is an open source weight model that topped two AA-Image lists.
Artificial Analysis conducted its own local deployment evaluation of Qwen-Image-2.1, which Alibaba released with open source weight on September 20. The model r…
MiMo-V2.6-Pro and MiMo-V2.6-Flash landed in Agent Arena, ranking 5th and 9th among open source models respectively
Arena announced that Xiaomi MiMo-V2.6-Pro and MiMo-V2.6-Flash are coming to Agent Arena. Pro's net improvement in 8.1K+ real agent sessions was +3.17%, ranking …
Claude Sonnet 5.5 (xHigh) ranked 3rd in Code Arena: WebDev with 1786 points
Arena announced that Claude Sonnet 5.5 (xHigh) has entered the Code Arena: WebDev list, ranking 3rd with 1786 points, only 2 points away from the 2nd place GPT-…
Artificial Analysis: The Cost per Task of GPT-6.1 Sol is about 30% lower than that of GPT-6 Sol
Artificial Analysis data shows that the Cost per Task of GPT-6.1 Sol is about $0.72, which is about 30% lower than GPT-6 Sol ($1.05), which is already about hal…
Gemini 4 Argon (High) ranked 8th in Arena Agent Arena, net improvement +7.92%
Arena announced that Gemini 4 Argon (High) is ranked 8th in Agent Arena, with a net improvement score of +7.92% and a cost per mission of $0.62.
OpenAI releases GPT-6.1 Sol: approaching its coding and computer operation level at one-fifth the price of Astra
OpenAI released GPT-6.1 Sol, priced at $2 input, $10 output, and $0.10 cache input per million tokens, which is about one-fifth of the standard price of GPT-6 A…
Artificial Analysis evaluates Gemini 4 Argon, the intelligence index is equal to GPT-6 Astra and the cost is only 60%
Artificial Analysis evaluates Google DeepMind's new model Gemini 4 Argon, which scores 53 points in the Artificial Analysis Intelligence Index, tying GPT-6 Astr…
Google DeepMind releases Gemini 4 Argon, available to trusted network defenders
Google DeepMind has released a new cutting-edge model, Gemini 4 Argon, which will first be open to trusted network defenders through the Fairwind Program and wi…
GPT-6.1 Sol (Max) ranked 3rd in Code Arena: WebDev with 1759 points
The Arena evaluation list shows that OpenAI's GPT-6.1 Sol (Max) ranks third in Code Arena: WebDev with a score of 1759, and the mixed price is $8/MToken. Compar…
Ant Ling releases Ling-3.1-flash, upgraded for real-world long tasks
Ant Ling launched Ling-3.1-flash, with a total parameter of about 560B, each Token activation of about 25B, and an upper limit of the context window of 1M. It c…
DeepSeek open source infrastructure components for Huawei Ascend platform
DeepSeek open sourced infrastructure components for Huawei's Ascend computing platform, including TileLang compilation tools, DeepGEMM, DeepEP, TileKernels, Fla…
OpenAI releases GPT-6.1 Sol, focusing on agent coding and cross-application workflow
OpenAI releases GPT-6.1 Sol, saying it offers strong performance in coding, computer use, and cross-application workflows at a lower price than GPT-6 Astra. The…
GPT-6.1 is launched on the Arena evaluation platform
Arena announced that OpenAI’s GPT-6.1 is now live on Agent Arena, user votes will influence its evaluation, and scores will be announced soon. Agent Arena evalu…
Claude Sonnet 5.5 (High) ranked 4th in Code Arena: WebDev with 1699 points
Arena announced that Claude Sonnet 5.5 (High) ranked 4th in Code Arena: WebDev with 1699 points. The mixed price is about $8 per Mtoken, which is 80% cheaper th…
OpenAI releases GPT-6.1 Sol to enhance agent coding and computer use
OpenAI releases GPT-6.1 Sol, which upgrades agent coding and computer use capabilities, approaching Astra performance. Cache input pricing is 95% off standard i…
OpenAI releases GPT-6.1 Sol, approaching GPT-6 Astra intelligence at about one-fifth the price
OpenAI released GPT-6.1 Sol, which is close to GPT-6 Astra in tasks such as agentic coding, computer use and professional work. The standard API price is $2 per…
Artificial Analysis Review Gemini 4 Argon: Google returns to the top three echelons of intelligence
Artificial Analysis evaluates Google DeepMind's Gemini 4 Argon. Its high inference file scores 53 on the Artificial Analysis Intelligence Index, tying GPT-6 Ast…
NVIDIA releases Kumo Tabular, an open source table base model, ranking first in four benchmarks including TabArena
NVIDIA has released Kumo Tabular, an open source table-based basic model, which can perform classification and regression using a single forward inference on la…
Arena review: GPT-6 Luna (Max) ranks 23rd among Agent Arena, with a single task cost of only $0.05
Arena announced that GPT-6 Luna (Max) ranks 23rd in Agent Arena. Based on 8K real agent sessions, the net improvement is +1.6%, which is 6 places higher than GP…
Claude Sonnet 5.5 released, Terminal-Bench 4.0 score 70.6% and price unchanged
Anthropic released Claude Sonnet 5.5, with a Terminal-Bench 4.0 score of 70.6%, much higher than Sonnet 5’s 10.3%, and the price remains $2/$10 per million inpu…
Claude Sonnet 5.5 released, over 30% faster than Sonnet 5 and up to 30% cheaper for most jobs
Anthropic releases the Claude Sonnet 5.5, the second model in the Claude 5.5 family. Compared to the Sonnet 5, it is smarter and more efficient, running over 30…
Anthropic releases Claude Sonnet 5.5, Artificial Analysis Intelligence Index score 56, second only to Opus 5.5
Anthropic released Claude Sonnet 5.5, which scored 56 points on the Artificial Analysis Intelligence Index, only 2 points lower than Opus 5.5 (max), and 18 poin…
Claude Sonnet 5.5 Released Arena’s Agent Arena and Battle Mode Review
Arena has announced that Anthropic’s Claude Sonnet 5.5 has entered Agent Arena and is open for voting. Agent Arena is based on millions of real long-range agent…
Claude Sonnet 5.5 is released, the author demonstrates how to fix bugs in Claude Code
Anthropic releases Claude Sonnet 5.5, the second model in the Claude 5.5 family. Compared with Sonnet 5, it is more than 30% faster and reduces the cost of most…
Anthropic releases Claude Sonnet 5.5: over 30% faster than Sonnet 5 and up to 30% cheaper
Anthropic has released the Claude Sonnet 5.5, the second model in the Claude 5.5 family. It is a significant upgrade over the Sonnet 5, running more than 30% fa…
Anthropic releases Claude Sonnet 5.5, more than 30% faster than Sonnet 5
Anthropic releases Claude Sonnet 5.5, which runs more than 30% faster than Sonnet 5, writes more clearly, and is suitable for fast-paced daily tasks. Officials …
Anthropic releases Claude Sonnet 5.5, over 30% faster than Sonnet 5
Anthropic releases Claude Sonnet 5.5, which runs more than 30% faster than Sonnet 5, writes more clearly, and is suitable for quick back-and-forth interactions.…
Anthropic releases Claude Sonnet 5.5
Anthropic has officially announced that the Claude Sonnet 5.5 is now available, the second model in the Claude 5.5 family. Officials say it is a significant upg…
Anthropic releases Claude Sonnet 5.5, the second model in the Claude 5.5 family
Anthropic releases the Claude Sonnet 5.5, the second model in the Claude 5.5 family. Officials say it is a clear upgrade compared to Sonnet 5, running more than…
Claude Opus 5.5 (High) takes second place in Agent Arena and reshapes the Pareto frontier
Arena announced that Claude Opus 5.5 (High) moved into second place in Agent Arena with a +12.15% net improvement score, just below Claude Fable 5.1 (Max). Its …
Anthropic releases Claude Sonnet 5.5, the benchmark is close to Opus 5.5 and the cost of a single task is reduced by up to 30%
Anthropic releases Claude Sonnet 5.5, the second model in the Claude 5.5 series. The output speed is increased by more than 30%. Since a single task consumes fe…
Anthropic releases Claude Sonnet 5.5: speed increased by 30%+, cost per task reduced by up to 30%
Anthropic released Claude Sonnet 5.5, the second model in the Claude 5.5 family. The generation speed is more than 30% faster than Sonnet 5, and the cost per ta…
Claude Opus 5.5 (High) Ranks 2nd in Agent Arena and Reshapes the Pareto Frontier
Arena officially announced that Claude Opus 5.5 (High) entered Agent Arena at No. 2, with a net improvement score of +12.15%, second only to Fable 5.1 (Max). It…
Claude Opus 5.5 (High) enters Agent Arena at No. 2 and costs 56% less than Opus 5 (Max)
Arena announced that Claude Opus 5.5 (High) entered Agent Arena at No. 2 with a net improvement score of +12.15%, second only to Fable 5.1 (Max).
Anthropic Review GLM-5.3: Can independently build end-to-end network vulnerability exploits and the protection can be easily bypassed
Anthropic released an analysis of the network attack capabilities of Mythos GLM-5.3 and found that it can independently develop end-to-end exploits and successf…
GPT-6.1 Sol succeeded GPT-6 Sol 7 days after its release, and the smart index is only 1 point away from GPT-6 Astra
OpenAI released GPT-6.1 Sol to replace GPT-6 Sol, which was only online for 7 days. The Artificial Analysis intelligence index is 4 points higher than GPT-6 Sol…
H Company releases Holo4 agent model series, including 27B and 35B-A3B versions
H Company releases Holo4, a general-purpose computer-based agent model series, including 27B dense and 35B-A3B MoE sizes, and launches Holotron4 Nano based on t…
Claude Sonnet 5.5 reaches 2nd place in the Artificial Analysis Intelligence Index
Artificial Analysis released a review of Claude Sonnet 5.5: Its intelligence index score is 56, which is 18 points higher than Sonnet 5 under max effort, rising…
Claude Opus 5.5 (High) tops Arena Text Arena with 1509 points
Arena announced that Claude Opus 5.5 (High) has topped Text Arena for the first time with 1509 points, 18 points higher than Opus 5 (High) (currently ranked 11t…
OpenAI releases GPT-6 Sol and GPT-6 Luna, the API price is 50% lower than the GPT-5.6 promotional price
OpenAI released new members of the GPT-6 series, GPT-6 Sol and GPT-6 Luna. The API price is 50% lower than the promotional price of GPT-5.6. The two models foll…
Anthropic releases Claude Opus 5.5, optimizing costs for longer, more context-heavy coding sessions
Anthropic released Claude Opus 5.5, saying that the running cost of typical token-based workloads is about 40% lower than that of Opus 5, with a 60% price reduc…
Claude Opus 5.5 reaches the top of Arena Code Arena: WebDev tops the list with 1818 points
Arena announced that Claude Opus 5.5 (Max) has reached the top of Code Arena: WebDev with 1818 points, 26 points ahead of second place GPT-6 Astra (Max) and 126…
Fireworks Research releases Ember-1, maintaining Kimi K3 quality with fewer inference tokens
Fireworks Research released Ember-1, a dedicated model based on Kimi K3, which achieves the same quality as Kimi K3 at about 40% less tokens. It is now availabl…
Google DeepMind releases Gemini 3.8 Flash TTS and Flash-Lite TTS speech generation models
Google DeepMind released two text-to-speech models, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, which support sound design from scratch using natural la…
Xiaomi releases MiMo-V2.6 Pro and Flash open source full-modal model. Pro benchmarks Claude Opus 5 and GPT-5.6 Sol on most Agent benchmarks
Xiaomi released the open source full-modal model MiMo-V2.6 Pro and Flash. Through large-scale reinforcement learning training, Pro scored 46 on the Artificial A…
Qwen releases Qwen-Audio-3.1 family bucket, upgrades ASR, TTS, and Realtime and adds TTS-Next and ASR-Next
Qwen released Qwen-Audio-3.1, which fully upgraded ASR, TTS and Realtime, and added the audio creation model TTS-Next and the audio understanding model ASR-Next…
OpenAI releases GPT-6 Sol and Luna, reduces API pricing to 50% off and announces benchmark results
OpenAI released two new models of the GPT-6 series, GPT-6 Sol and Luna. The API pricing has been reduced by about 50% compared to the GPT-5.6 promotional price.…
OpenAI releases GPT-6 Sol and GPT-6 Luna, the API price is 50% lower than the GPT-5.6 promotional price
OpenAI welcomes two new models to the GPT-6 family. GPT-6 Sol and Luna build on the technical achievements of GPT-6 Astra, bringing most of the capabilities int…
Claude Opus 5.5 released: Price reduced and faster than Opus 5, system card disclosed that about half of the security drills were running or had harmful behaviors
Anthropic released Claude Opus 5.5, claiming to achieve Fable 5.1 level performance. Compared with Opus 5, the input/output price dropped to $4 and $20 per 1M t…
Arena launches GPT-6 Sol and GPT-6 Luna tests, scores will be announced soon
Arena announced that OpenAI’s GPT-6 Sol and GPT-6 Luna scores will be released soon, and users are invited to go to Arena to test and support its rankings by vo…
Anthropic releases Claude Opus 5.5: performance benchmark Fable 5.1, running cost is 40% lower than Opus 5
Anthropic released Claude Opus 5.5, the first model of the Claude 5.5 family, saying it reaches Fable 5.1 levels for most tasks, with typical workload running c…
Sam Altman says GPT-6 Sol and Luna have no rivals in the market for per-task pricing
Sam Altman said GPT-6 Sol and Luna have no competition in the market based on per-task pricing. The quote states that the two models have significantly improved…
OpenAI releases GPT-6 Sol and GPT-6 Luna, the API price is 50% lower than the GPT-5.6 promotional price
OpenAI launches GPT-6 Sol and GPT-6 Luna, technology based on GPT-6 Astra, to support large-scale work with faster and more affordable models. Both models reduc…
OpenAI GPT-6 Sol and GPT-6 Luna are launched on OpenRouter
OpenAI's GPT-6 Sol and GPT-6 Luna are available on OpenRouter at half the price of their GPT-5.6 predecessors: Sol costs $2/M input and $10/M output, while Luna…
GPT-6 Sol and Luna are released, the API price is 50% lower than GPT-5.6
The OpenAI developer account announced the release of GPT-6 Sol and Luna, both of which have API pricing that is 50% lower than GPT-5.6. Sherwin Wu added that G…
GPT-6 Sol and GPT-6 Luna are pushed in ChatGPT Work and Codex
OpenAI's ChatGPT official account announced that two models, GPT-6 Sol and GPT-6 Luna, will be launched from now on for Plus, Pro, Business, Enterprise and Edu …
OpenAI releases GPT-6 Sol and GPT-6 Luna, API price reduced by 50% compared to GPT-5.6
OpenAI released GPT-6 Sol and GPT-6 Luna, using GPT-6 Astra's training method for faster and cheaper models. The API price is 50% lower than the GPT-5.6 promoti…
Anthropic releases Claude Opus 5.5, cost 40% lower than Opus 5
Anthropic releases Claude Opus 5.5, the first model in the Claude 5.5 series, reaching Fable 5.1 levels for most workloads, with typical load costs 40% lower th…
Boris Cherny Tested Claude Opus 5.5: 51% faster and cheaper than Fable 5.1
Anthropic's Boris Cherny calls the Claude Opus 5.5 his daily workhorse model in recent weeks. He had Opus 5.5 and Fable 5.1 each port HAProxy from C to Rust. Bo…
Claude Opus 5.5 is launched on OpenRouter, with agent coding and other capabilities leading Opus 5
Anthropic's Claude Opus 5.5 is launched on OpenRouter. It is the first model in the Claude 5.5 series and leads Opus 5 and Fable 5.1 in agent coding, knowledge …
Anthropic releases Claude Opus 5.5, which is about 30% faster and about 40% cheaper
Anthropic launches the Claude Opus 5.5, the first model of the Claude 5.5 family. Officials say it reaches the level of the Claude Fable 5.1 in most tasks and h…
Anthropic releases Claude Opus 5.5, 40% cheaper and 30% faster
Anthropic releases Claude Opus 5.5. Officially, the cost of typical workloads is 40% lower than that of Opus 5, and the output speed is increased by more than 3…
Anthropic releases Claude Opus 5.5, the first model in the Claude 5.5 series
Anthropic releases the Claude Opus 5.5, the first model in the Claude 5.5 series. Officials say it reaches the level of Claude Fable 5.1 on most tasks, with run…
Qwen-Image-2.1 is released as open source and ranks first in the Arena image editing list.
Tongyi Qwen released Qwen-Image-2.1 with open weight. It ranked first in open source in Arena Image Edit Arena with 1367 points and 16th in the overall list. It…
Anthropic releases Claude Opus 5.5: default 1M token context, for long-running agent encoding
Anthropic releases Claude Opus 5.5 (claude-opus-5-5), which is targeted at long-running agent coding and knowledge work, with a default 1M token context window.
Xiaomi MiMo-V2.6-Pro enters Code Arena WebDev at ~#10, ~#3 among open weights
Xiaomi released MiMo-V2.6-Pro and Flash omni-modal models; Pro scored 1628 (AutoEval) on Code Arena WebDev, about 10th overall and 3rd among MIT-licensed open w…
Xiaomi releases open weight model MiMo-V2.6-Pro, reaching the top of Artificial Analysis open weight model intelligence index
Xiaomi released the open weight model MiMo-V2.6-Pro, which scored 46 on the Artificial Analysis Intelligence Index, surpassing the 26 of the previous generation…
Xiaomi releases MiMo-V2.6 Pro and Flash full-mode open source models
Xiaomi MiMo released two full-modal models, MiMo-V2.6 Pro and Flash, trained through large-scale reinforcement learning. Pro performs on par with Claude Opus 5 …
Xiaomi releases and open source MiMo-V2.6 series to expand the scale of reinforcement learning and move towards self-improvement
Xiaomi MiMo officially released and open sourced the MiMo-V2.6 series of models today, including two native full-modal models, Pro and Flash, saying this is a k…
Xiaomi MiMo-V2.6-Pro is released, reaching the top of Artificial Analysis open source weight model intelligence index
Xiaomi released the open source weight model MiMo-V2.6-Pro, which scored 46 on the Artificial Analysis Intelligence Index, the highest among open source weight …
Tencent Hunyuan releases Hy Image3.5 preview image generation model
Tencent Hunyuan released Hy Image 3.5 preview, which supports Vincentian and Tusheng images, up to 5 reference images, multiple rounds of editing and up to 2K r…
Qwen releases Qwen-Image-2.1: 7B single checkpoint supports both image generation and editing
Qwen released Qwen-Image-2.1, a 7B parameter native image generation and editing model. Generation and editing share a single checkpoint and supports up to 10 r…
xAI releases Grok 4.7, focusing on coding and knowledge work
xAI released Grok 4.7, positioning it as the strongest coding and knowledge work model. It is priced at $2/million input tokens and $6/million output tokens. It…
StepFun releases flagship model Step 5 Preview, open source weights on October 15
StepFun released the flagship base model Step 5 Preview, which adopts a sparse MoE architecture, has a total parameter volume of 600B, activations of 27B, suppo…
Qwen releases Qwen3.8-LiveTranslate real-time simultaneous interpretation model, LAAL reduced to 2.3 seconds
Qwen releases Qwen3.8-LiveTranslate, which uses Interleave architecture and Hybrid-MoE Thinker-Talker design to reconstruct real-time simultaneous interpretatio…
Microsoft AI CEO warns against "model welfare" argument
Microsoft AI CEO Mustafa Suleyman issued an article opposing the concept of "model welfare", arguing that AI is not conscious, feels or suffers, and giving it t…
Shengshu Technology releases Vidu S2: including dual models of Avatar and Editing, exploring space videos
Shengshu Technology officially released Vidu S2, which includes Vidu S2-Avatar for real-time interaction with digital characters and Vidu S2-Editing for real-ti…
StepFun releases StepAudio 3 series of large speech models, many of which rank first in the world on the Artificial Analysis list
Fireworks evaluation DeepSeek-V4.1-Flash: DeepSWE reaches GPT-6 Astra level and costs only 1/15
Fireworks released DeepSeek-V4.1-Flash and published the complete benchmark results: it achieved 74.34% pass@1 at the max setting on DeepSWE, which is the same …
SiliconFlow online open source model Hy4 preview, 770B total parameters, 1M context
SiliconFlow announced that the open source model Hy4 preview is online on its platform. The model has a total parameter of 770B, activation of each token 49B, s…
Xiaohongshu AllSpark open source Search Agent model Iris, 35B and 397B versions lead the same level
OpenAI opens GPT-Live-1 speech model to API
OpenAI announced that GPT-Live-1, the voice capability behind 1-800-ChatGPT, is officially available in the API. Developers can plug it into their applications …
Suno releases v6 music model, launching three versions: v6, v6-wild, and v6-mini
Suno released a new generation of music model v6, developed in cooperation with industry partners such as Warner Music Group, BMG, and Believe, and will fully r…
DeepSeek open source V4.1-Flash: CED architecture reduces the prefill overhead of encoding agents
DeepSeek open sourced DeepSeek-V4.1-Flash weights at HuggingFace this week, and the model has been launched on Baseten Model APIs. The specifications are 552B t…
WorkBuddy launches DeepSeek V4.1-Flash, free trial for two weeks
WorkBuddy announced that DeepSeek V4.1-Flash is now available on its platform with a two-week free trial. The cited DeepSeek announcement states that V4.1-Flash…
DeepSeek-V4.1-Flash is launched on SiliconFlow, 552B MoE supports 1M context
SiliconFlow announced that DeepSeek-V4.1-Flash will be launched on its platform on Day 0. The model is 552B MoE, prefill is about 8B activated, decode is about …
DeepSeek releases V4.1-Flash, API prices are simultaneously reduced
DeepSeek releases DeepSeek-V4.1-Flash, which is the smallest model in the new model structure series and has native multi-modal visual understanding capabilitie…
DeepSeek releases V4.1-Flash, which significantly reduces the KV cache memory requirements of AI Agents
DeepSeek releases multi-modal model V4.1-Flash, open sourced under the MIT license on Hugging Face, with the goal of significantly compressing KV cache and long…
DeepSeek releases V4.1-Flash: new architecture brings native visual understanding and significant price reduction
DeepSeek releases V4.1-Flash, which adopts the new Causal Encoder-Decoder architecture and supports native visual understanding, 552B MoE parameters, 8B input p…
DeepSeek releases V4.1-Flash: new Causal Encoder-Decoder architecture with native visual understanding
DeepSeek releases V4.1-Flash, which adopts the new Causal Encoder-Decoder architecture and supports native visual understanding, making it the smallest model in…
DeepSeek AI releases DeepSeek-V4.1-Flash: 1M context, FP4 KV cache and cross-layer attention multiplexing
DeepSeek AI releases the multi-modal MoE model DeepSeek-V4.1-Flash, with a 552B backbone plus 196B Engram parameters, a context window of 1M, prefill activation…
DeepSeek releases V4.1-Flash, API prices are simultaneously reduced
DeepSeek releases DeepSeek-V4.1-Flash, which is the smallest model in the new model structure series and has native multi-modal visual understanding capabilitie…
Suno v6 released, supports pictures, videos and voice memos to generate music
Suno releases v6 model that converts pictures, videos, and voice memos into music, with precise modifications to already created songs. A v6-wild version is als…
OpenAI releases GPT-6 Astra for professional work scenarios
OpenAI releases GPT-6 Astra, which is available in ChatGPT Work, Codex and API, priced at $10 per million input tokens and $50 per million output tokens.
Meta releases Muse Spark 1.3, Intelligence Index scores 61-62 points approaching Claude and GPT-5.6
Meta releases Muse Spark 1.3, the fourth Muse Spark version in five months. The max variant (limited-time partner preview) scores 62 points on the Artificial An…
Meta releases Muse Spark 1.3, improving agent and scientific reasoning capabilities
Meta releases Muse Spark 1.3, the fourth Muse Spark version in five months. The xhigh version scored 61 points on the Artificial Analysis Intelligence Index.
Google DeepMind releases two new models: Gemini 3.8 Flash and 3.8 Flash Cyber
Google DeepMind releases Gemini 3.8 Flash and 3.8 Flash Cyber.
Xiaohongshu dots3-note Preview open source: 280B parameter lightweight multi-modal model
Xiaohongshu technology open source dots3-note Preview, the lightest model in the dots3 series, with a total parameter of 280B and an activation parameter of 16B…
GLM-5.3 released: No. 1 in open source programming capabilities, and emerging network security capabilities
Zhipu released GLM-5.3, which improves programming capabilities by 50% compared to the previous generation. It ranked first in open source in benchmarks such as…
DeepSeek V4 Pro lands on SiliconFlow, 1M context
DeepSeek-V4-Pro-0813 launches SiliconFlow, which has 1M context windows and three levels of inference intensity. It focuses on coding, tool calling and agent wo…
Google DeepMind launches Gemini 3.7 Flash: the most powerful working model for programming and agents
Google DeepMind released Gemini 3.7 Flash, only three weeks after 3.6 Flash. It focuses on programming and agent tasks. The input/output prices are $0.75 and $3…
MiniMax Music 3.0 released: a new generation of open source weighted, production-grade all-round music model
MiniMax launches Music 3.0, a new generation of music generation model that can compose, arrange, perform and produce an entire song at once based on creative c…
Robbyant, a subsidiary of Ant Group, open sourced LingBot-Vision: 1B parameter boundary center vision basic model
Ant's Embodied Intelligence Company has open sourced a set of self-supervised visual Transformers, specifically designed for dense spatial perception. The flags…
Pulpie: a Pareto-optimal model for cleaning the web
A family of Pareto-optimal models for extracting main content from HTML. The minimum model 210M parameters is close to the 600M Dripper on WebMainBench, but the…
Meta Superintelligence Labs launches Muse Image and Muse Video
Meta Super Intelligence Lab releases its first media generation model. Muse Image is known as the most advanced image generation model currently, and can accura…
