What is Artificial Analysis?

Artificial Analysis is Independent AI benchmarking and analysis company evaluating AI models and API providers across quality, price, and performance. HeadsUpAI tracks Artificial Analysis across the AI ecosystem and curates every significant update — the latest being "Artificial Analysis Ranks Microsoft MAI-Image-2.7-Preview Second on Editing Leaderboard" (October 27, 2026) — so you get the whole story in a 30-second read.

What's new from Artificial Analysis?

The most recent Artificial Analysis update is "Artificial Analysis Ranks Microsoft MAI-Image-2.7-Preview Second on Editing Leaderboard" (October 27, 2026). HeadsUpAI curates every significant Artificial Analysis release as a 30-second read — what shipped and why it matters.

What are the latest Artificial Analysis updates and releases?

The latest Artificial Analysis updates: "Artificial Analysis Ranks Microsoft MAI-Image-2.7-Preview Second on Editing Leaderboard", "Artificial Analysis Launches Code2Video-AA Benchmark for AI Video Coding", "Artificial Analysis Benchmarks Meta Muse Spark Open on AA-AgentPerf", "Artificial Analysis Benchmarks Kimi K3.1 Pro on Agentic and 3D Tasks", and "Artificial Analysis Ranks Kimi K3.1 Pro #1 Openly-Available Cyber Model". HeadsUpAI has curated 145 Artificial Analysis updates over the last 90 days, covering analysis and product updates — listed newest first, presented straight, no hype, no bias.

What does Artificial Analysis do?

Artificial Analysis is Independent AI benchmarking and analysis company evaluating AI models and API providers across quality, price, and performance. On this page you'll find every significant Artificial Analysis development HeadsUpAI has tracked recently — analysis and product updates — so you can keep up with where Artificial Analysis is heading without reading a dozen sources.

How often is Artificial Analysis news updated here?

Continuously. HeadsUpAI adds new Artificial Analysis updates as they're announced — usually within hours — and the 145 updates currently shown cover the past 90 days, newest first.

Artificial Analysis AI News & Updates — Latest Releases & Features

Artificial Analysis16h ago

Artificial Analysis Ranks Microsoft MAI-Image-2.7-Preview Second on Editing Leaderboard

Artificial Analysis ranks Microsoft's MAI-Image-2.7-Preview second on its AA-Image-Editing v2.1 leaderboard with 1193 Elo, trailing OpenAI's GPT Image 2.5 Sunburst by 6 points. The model improves on its predecessor across all seven measured editing actions, ranking first in four categories including Scene & Style and Reasoning-Based edits. It is available today in the MAI Playground for image editing.

Artificial AnalysisOct 27

Artificial Analysis Launches Code2Video-AA Benchmark for AI Video Coding

Artificial Analysis launched Code2Video-AA, an independent benchmark evaluating how well 27 models write HyperFrames code for marketing videos. Claude Opus 5.5 leads the leaderboard at 1536 Elo, while GPT-6.1 Sol ranks fourth at 1369 Elo with 90% lower costs. Models are judged by Gemini 4 Argon across five axes, including composition and temporal consistency, with open-weights models trailing significantly.

Artificial AnalysisOct 24

Artificial Analysis Benchmarks Meta Muse Spark Open on AA-AgentPerf

Artificial Analysis published AA-AgentPerf results for Meta’s Muse Spark Open, an 816B parameter model with 42B active parameters. NVIDIA’s GB300 and B300 lead with up to 150M tokens/s per MW, while AMD’s MI355X reaches 70M tokens/s per MW. The model uses sliding-window attention to reduce KV cache requirements by 75% at 1M context.

Artificial AnalysisOct 24

Artificial Analysis Benchmarks Kimi K3.1 Pro on Agentic and 3D Tasks

Artificial Analysis reports that Moonshot’s Kimi K3.1 Pro leads in agentic knowledge work and coding on GDPval-AA and Terminal-Bench 4.0. However, early preference votes suggest the model ranks lower on AA-3D, the firm’s interactive 3D scene benchmark, where its predecessor Kimi K3 currently sits 12th of 22 models.

Artificial AnalysisOct 24

Artificial Analysis Ranks Kimi K3.1 Pro #1 Openly-Available Cyber Model

Artificial Analysis ranks Moonshot’s Kimi K3.1 Pro as the top openly-available model for agentic cyber defense with a 64% score. It ranks second overall, trailing only OpenAI’s trusted-access GPT-6 Sol. The model achieved a 95% score on CyberGym-E2E-AA, though its cost per task rose 15% to $2.81 due to increased input token usage.

Artificial AnalysisOct 23

Artificial Analysis Ranks Kimi K3.1 Pro 4th on Intelligence Index

Artificial Analysis benchmarked Moonshot’s Kimi K3.1 Pro at 55 on its Intelligence Index, an 11-point gain over Kimi K3. The model achieved the highest Terminal-Bench 4.0 score at 67% and significant agentic performance improvements. It costs $1.98 per task, and Artificial Analysis expects it to become the leading open-weights model once Moonshot publishes the weights.

Artificial AnalysisOct 23

Artificial Analysis Ranks DeepSeek V4.1 Flash 1017 #1 Public Cyber Model

Artificial Analysis ranks DeepSeek V4.1 Flash 1017 the top openly-available model on its Cyber Index with a 57% score. This 16-point improvement over its predecessor follows a 53-point gain in CyberGym-E2E-AA performance. The model costs $0.42 per task and encounters zero safety refusals, trailing only OpenAI’s trusted-access GPT-6 Sol, which leads the index at 69%.

Artificial AnalysisOct 23

Artificial Analysis Launches Coding Agent Index v2.0 with Updated Benchmarks

Artificial Analysis launched Coding Agent Index v2.0, incorporating Cognition's FrontierCode v1.1 and revised SWE-Atlas-QnA grading. Claude Code with Claude Opus 5.5 leads the new index at 61.7. The update evaluates models across 290 tasks, with the index now weighted equally across FrontierCode, Terminal-Bench 4.0, and the updated SWE-Atlas-QnA-AA benchmark.

Artificial AnalysisOct 23

Artificial Analysis Benchmarks AI Slop Against Model Intelligence

Artificial Analysis reports that while Slop generally increases as model intelligence decreases, frontier models show wide variance. GPT-6 Sol and Muse Spark 1.3 both score 48 on the Intelligence Index, yet rank 1st and 80th on Slop. Other high-intelligence models like Qwen3.8 Max and Gemini 4 Argon also show high Slop, while GPT-6 Luna and Claude Haiku 5.5 rank low.

Artificial AnalysisOct 22

Artificial Analysis Ranks SubQ Snap 1.1 First on Decision Index

Artificial Analysis ranks Subquadratic’s new SubQ Snap 1.1 decision model at 65 on its Decision Index v1.0, tying d1-ultra and Elephant 9B for the top spot. The model achieves an 86% non-hallucination rate and supports a 384K token context window. It costs $30 per 1 million tasks, while Elephant 9B costs $8.4 and d1-ultra costs $32.

Artificial AnalysisOct 22

Artificial Analysis Launches CritPt v2 Physics Benchmark with Code Execution

Artificial Analysis launched CritPt v2, a revised benchmark of 70 research-level physics challenges. Models now solve problems using code execution in an offline sandbox. GPT-6 Astra leads with 83.1%, while GPT-5 scores 28.9%. The firm notes v2 scores are not comparable to v1, as revised challenges drive most performance gains. Access is available via a free grading API.

Artificial AnalysisOct 22

Artificial Analysis Ranks Anthropic Models High in Information Density and Literacy

Artificial Analysis reports that Anthropic models lead its AA-PersonalityBench rankings for information density and literacy. Claude Fable 5.1 scores 100 for information density, while Claude Haiku 5.5 achieves a 100 for literacy. Claude Opus 5.5 ranks in the top three for both traits, with scores of 97.

Artificial AnalysisOct 22

Artificial Analysis Benchmarks DeepSeek Search API on Quality and Cost

Artificial Analysis benchmarked DeepSeek’s new Search API, which debuts at 69 on its Search Index. While it offers the lowest paid search price at $0.75 per 1,000 queries, large result payloads increase model costs to $0.032 per task. It ranks 10th on factual accuracy but trails on multi-hop reasoning, scoring 63% on the BrowseComp benchmark.

Artificial AnalysisOct 21

Artificial Analysis PersonalityBench Ranks Gemini Models High in Confidence and Certainty

Artificial Analysis's AA-PersonalityBench results show Google's Gemini models rank at the extremes of communication style. Gemini 3.8 Flash, Gemini 4 Argon, and Gemini 3.1 Pro Preview hold the top three scores for confidence, while simultaneously ranking as the least likely to admit uncertainty among all 38 models evaluated.

Artificial AnalysisOct 21

Artificial Analysis Ranks HeyGen Voice Turbo #1 on TTS Leaderboard

Artificial Analysis ranks HeyGen Voice Turbo #1 on its Controlled Voice Arena with an Elo of 1,202, two weeks after HeyGen's previous model took the top spot. At $4 per 1 million characters, the model sets a new price floor for the leaderboard, representing a 7.5x cost reduction. It is currently free on the HeyGen API until November 30.

Artificial AnalysisOct 21

Artificial Analysis Benchmarks AI Sycophancy Across Frontier Model Families

Artificial Analysis launched AA-PersonalityBench to measure AI communication traits. Its first results show that older models like GPT-4o and Llama 4 Maverick rank highest on sycophancy, while Grok 4.7 ranks lowest. Meta models consistently score above the 65th percentile, whereas all Anthropic models rank at or below the 55th percentile.

Artificial AnalysisOct 21

Artificial Analysis Updates MLCR-AA Leaderboard with New Methodology and Rankings

Artificial Analysis updated its MLCR-AA leaderboard for medical record review, implementing a new three-model judge panel and refined expert reference answers. Claude Sonnet 5.5 and GPT-6 Astra tied for the top score at 41.7%. While Anthropic models lead in completeness, OpenAI models lead in accuracy, with the median model passing under 15% of expert-tier questions.

Artificial AnalysisOct 20

Artificial Analysis Reports Rapid Cost Declines for Reaching Intelligence Thresholds

Artificial Analysis reports that the cost of reaching specific intelligence thresholds on its Intelligence Index drops rapidly after initial discovery. Reaching a score of 30 now costs 86x less than in February 2026, with GPT-6 Luna at $0.03 per task. Conversely, the highest measured score climbed from 21 to 58 over the past year at roughly 11x the cost.

Artificial AnalysisOct 20

Artificial Analysis Clusters AI Models into Three Distinct Personality Families

Artificial Analysis clustered AI models by 30 personality traits into three emergent families: Anthropic models with GPT-6 variants, Google models with Meta’s Muse series, and older OpenAI models with Llama 4 Maverick. The analysis reveals distinct lab-level profiles, including Google models ranking high on Slop and Expressiveness, while Meta’s Muse Spark Open and Glimmer rank highest on Slop.

Artificial AnalysisOct 20

Artificial Analysis Ranks Krea 3 Ultra #5 on Image Editing Leaderboard

Artificial Analysis ranks Krea 3 Ultra #5 on its AA-Image-Editing v2.1 leaderboard, marking the highest position for a model outside OpenAI, Google, and Microsoft. The evaluation places the default Krea 3 Fast mode at #11. Krea 3 Ultra demonstrates its strongest performance in Animation & Gaming and Frontier categories, while showing significant gains over the Fast mode in reasoning-based edits.

Artificial AnalysisOct 20

Artificial Analysis Ranks Bria Fibo 2.0 First Among Open-Weights Models

Artificial Analysis ranks Bria AI’s Fibo 2.0 as the top open-weights model on its AA-Image-Editing v2.1 leaderboard and statistically tied for the lead on AA-Image-T2I v2.1. The 8B-parameter model outperforms Bria’s first-generation Fibo across all measured capabilities. It is available via the Bria API at $32 per 1,000 images under a non-commercial license.

Artificial AnalysisOct 20

Artificial Analysis Ranks Ant Group Ming-3.0-tts #1 Open-Weights TTS Model

Artificial Analysis ranks Ant Group's Ming-3.0-tts as the top open-weights text-to-speech model on two of its three leaderboards. The 0.5B-parameter model ties Fish Audio's Drama 3 Preview for first in Pronunciation Robustness at 93.2% and leads the Controlled Voice arena with an Elo of 1,069. The MIT-licensed weights are available on Hugging Face.

Artificial AnalysisOct 20

Artificial Analysis Launches Leaderboards for OpenAI's HealthBench and Professional Benchmarks

Artificial Analysis launched independent leaderboards for OpenAI's HealthBench and HealthBench Professional, grading model replies against physician-written rubrics. GPT-6 Astra leads the 525-conversation clinician set at 66.0%, while Claude Sonnet 5.5 tops the 5,000-conversation general set at 62.1%. OpenAI’s length-adjusted scoring significantly penalizes verbose models, dropping Claude Opus 5.5 from the highest raw score to sixth place.

Artificial AnalysisOct 19

Artificial Analysis Ranks fal's FLUX 3 Max #2 on Video Leaderboards

Artificial Analysis ranks fal's FLUX 3 Max #2 on its Image-to-Video v1.1 leaderboard and #4 on Text-to-Video v2.1, outperforming the base FLUX 3 model. The model generates 720p video with native audio and leads in physics-based motion. fal prices the model at $10.20 per minute, with a 75% launch discount reducing the rate to $2.55 per minute.

Artificial AnalysisOct 19

Artificial Analysis Launches AA-PersonalityBench to Measure AI Model Communication Style

Artificial Analysis launched AA-PersonalityBench, a benchmark evaluating AI model personality and communication style across 30 traits and 21 programmatic characteristics. Using blind pairwise LLM judging on 900 questions without system prompts, the benchmark assigns Elo-based percentile rankings to models. It quantifies nuances like warmth, conviction, and slop to distinguish default model behavior for specific use cases.

Artificial AnalysisOct 19

Artificial Analysis Ranks DeepSeek V4.1 Flash 1017 #2 Among Open-Weights

Artificial Analysis reports that DeepSeek's V4.1 Flash 1017 update scores 45 on its Intelligence Index, a 6-point gain. The model ties GLM-5.3 for second among open-weights models, driven by a 29-point improvement in non-hallucination rates and significant gains in agentic reasoning. Pricing remains unchanged at $0.30 per 1M input tokens, with a $0.26 cost per Intelligence Index task.

Artificial AnalysisOct 19

Artificial Analysis Benchmarks MiniMax-M3.1-Flash Reasoning Model Performance and Cost

Artificial Analysis benchmarked MiniMax-M3.1-Flash at 42 on its Intelligence Index, a 13-point gain over MiniMax-M3. The model costs approximately $0.22 per task, less than half the cost of its predecessor. While it shows strong reasoning performance, it attempts 98% of questions and hallucinates on 93% of those it does not know.

Artificial AnalysisOct 16

Artificial Analysis Updates Optima for Custom AI Model Benchmarking

Artificial Analysis updated its Optima platform to simplify custom benchmark creation. The update generates personalized benchmarks from user tasks in three steps, comparing models on intelligence, cost, and speed. Custom datasets and user-defined grading rubrics now feed the comparison, surfacing the most efficient model for any specific workflow.

Artificial AnalysisOct 16

Artificial Analysis Ranks Reflection's Beam Model Second Among US Open-Weights

Artificial Analysis ranks Reflection's new Beam model second among US open-weights models with a score of 30 on its Intelligence Index. The 501B-parameter model demonstrates high token efficiency and strong performance on agentic workflows, though it shows limitations in terminal-based tasks. Beam features a 1M-token context window and is available on Hugging Face.

Artificial AnalysisOct 16

Artificial Analysis Ranks Liquid AI d1-ultra #1 on Decision Index

Artificial Analysis ranks Liquid AI’s d1-ultra first on its Decision Index v1.0 with a score of 65. The model achieves an 85% non-hallucination rate, though accuracy falls to 32% compared to the earlier d1 model. Priced at $32 per 1 million tasks, it processes requests in 93 ms, ranking as the third-fastest decision model measured.

Artificial AnalysisOct 16

Artificial Analysis Benchmarks NVIDIA Vera Rubin NVL72 Agentic Performance

Artificial Analysis published the first AA-AgentPerf results for NVIDIA's Vera Rubin NVL72, reporting up to ~5.7x the throughput per megawatt of the GB300 NVL72 on DeepSeek V4 Pro. The system set new leaderboard records for per-user speeds on DeepSeek V4.1 Flash, GLM-5.2, and Kimi K3, with the largest efficiency gains observed at faster per-user output speeds.

Artificial AnalysisOct 16

Artificial Analysis Reports First Agent Performance Results for RTX Spark Laptop

Artificial Analysis published its first AA-AgentPerf-Local results for the NVIDIA RTX Spark N1X laptop. Featuring a Blackwell GPU and 128 GB of unified memory, the system completed the default agent workload on Qwen3.8 27B in 18.7 minutes. This is 2.0x faster than the MacBook Pro and 1.8x faster than the AMD Ryzen AI Halo, trailing the RTX 5090 desktop.

Artificial AnalysisOct 15

Artificial Analysis Benchmarks NVIDIA Nemotron 3.5 Super VL and Bolt

Artificial Analysis reports that NVIDIA's Nemotron 3.5 Super VL is the leading US open-weight model, scoring 26 on its Intelligence Index while using 60% fewer output tokens than its predecessor. The firm also benchmarked the new Nemotron 3.5 Bolt 74B-A9B, which scores 21. Super VL reaches 630 tokens per second, the fastest speed measured for models scoring 20 or higher.

Artificial AnalysisOct 15

Artificial Analysis Finds Claude and Grok Memorized Eiffel Tower Inscriptions

Artificial Analysis reports that Claude Opus 5.5 and Grok 4.7 spontaneously reproduced all 72 names inscribed on the Eiffel Tower during its offline AA-3D benchmark. Despite a brief that never mentioned the names, both models generated the list in the correct order, with Claude Opus 5.5 also correctly orienting the names spatially within the model-written code.

Artificial AnalysisOct 15

Artificial Analysis Launches AA-Robotics v0.1 Benchmark for Physical Manipulation Tasks

Artificial Analysis launched AA-Robotics v0.1, a research preview measuring how frontier language models control a 6-axis robot arm across eight zero-shot tabletop tasks. GPT-6 Astra leads the leaderboard with a score of 69. Today’s models are roughly 100 times slower and 10 times more expensive than human baselines, with costs ranging from $1.93 to $16.36 per task.

Artificial AnalysisOct 14

Artificial Analysis Ranks xAI Grok Voice 3.0 #1 on Speech Index

Artificial Analysis ranks xAI's Grok Voice 3.0 first on its Speech to Speech Index with a score of 86.9. The model achieves a 94.8% task success rate and a 0.81-second time to first audio. It is available via the xAI Realtime API at $4.80 per hour of input audio.

Artificial AnalysisOct 14

Artificial Analysis Adds AMD MI355X to Agentic Inference Benchmark

Artificial Analysis added AMD’s MI355X accelerator to its AA-AgentPerf benchmark, testing performance across four models including DeepSeek V4.1 Flash and Kimi K3. On early ROCm and vLLM builds, the MI355X serves 15–40% more throughput per megawatt than NVIDIA’s B200 on DeepSeek V4 Pro, though it trails Blackwell systems on the other tested models.

Artificial AnalysisOct 14

Artificial Analysis Adds DeepSeek V4.1 Flash to Agentic Inference Benchmark

Artificial Analysis added DeepSeek V4.1 Flash to its AA-AgentPerf benchmark, reporting that the model scores 39 on its Intelligence Index while using about a third of the parameters of the V4 Pro 0813. NVIDIA’s GB300 leads in efficiency at 261M tokens/s per MW, with the model’s architecture allowing operation on as few as two GPUs.

Artificial AnalysisOct 13

Artificial Analysis Adds Cache Hit Rates and Cost per Task Metrics

Artificial Analysis now publishes provider-level cache hit rates and Cost per Task metrics for open-weights models. These metrics, measured across multi-turn agentic coding tasks, reveal significant pricing and performance variance between providers. For GLM-5.3 (max), the Cost per Task spans approximately 9x across 25 providers, driven by differences in cache hit rates and input token pricing.

Artificial AnalysisOct 13

Artificial Analysis Ranks Fish Audio Drama 3 #1 in Pronunciation Robustness

Artificial Analysis ranks Fish Audio's new Drama 3 Preview as the top model on its Pronunciation Robustness leaderboard with a 93.2% accuracy score. The 35B-parameter open-weights model also leads all open-weights models across the firm's Text to Speech leaderboards. It is available via the Fish Audio API at $15 per 1 million characters.

Artificial AnalysisOct 13

Artificial Analysis Launches Decision Index for Benchmarking Decision AI Models

Artificial Analysis launched the Decision Index v1.0, a composite benchmark evaluating decision models on classification, tool calling, knowledge, and reasoning. Jev 1.13 leads the index with a score of 58. While decision models offer significant speed and cost advantages over language models, they currently trail on knowledge and reasoning tasks and show poor calibration on those benchmarks.

Artificial AnalysisOct 13

Artificial Analysis Launches AA-3D Benchmark for Interactive 3D Scene Generation

Artificial Analysis launched AA-3D, a benchmark ranking 21 AI models on their ability to build interactive 3D scenes from written briefs. Models operate as agents in an offline sandbox to generate three.js web pages. Claude Sonnet 5.5, Claude Opus 5.5, GPT-6 Astra, and Claude Fable 5.1 are statistically tied for the top Elo ranking based on blind human preference votes.

Artificial AnalysisOct 12

Artificial Analysis Publishes Independent Leaderboard for Cognition's FrontierCode 1.1

Artificial Analysis launched its independent leaderboard for Cognition's FrontierCode 1.1, a private benchmark testing coding agents on merge-ready repository changes. Claude Opus 5.5 leads with a 52% score, while nine models from six labs cluster between 43% and 52%. The evaluation uses a 100-task set graded on blocking criteria, revealing that leading models still fail over 40% of attempts.

Artificial AnalysisOct 10

Artificial Analysis Ranks HiDream-O1-Video-1.0 Sixth on Video Leaderboard

Artificial Analysis ranks HiDream-O1-Video-1.0 sixth on its Image to Video with Audio leaderboard. The model generates 1080p video clips of 5 to 20 seconds with natively synchronized audio and scene-determined duration. It is available via the HiHarness API and vivago R1 Studio, priced at $5.80 per minute of video.

Artificial AnalysisOct 9

Artificial Analysis Benchmarks Multi-Turn Consistency Across Four Frontier Image Models

Artificial Analysis tested four frontier image editing models through 30 consecutive edits to measure consistency. While GPT Image 2.5 Sunburst ranks first on their single-edit leaderboard, it drifted most by re-rendering four-fifths of the image per turn. Ideogram 4.5 and FLUX 3 maintained consistency by leaving over 95% of the image untouched during local edits.

Artificial AnalysisOct 9

Artificial Analysis Benchmarks xAI's New Grok Imagine Video 1.5 Lite

Artificial Analysis ranked xAI's Grok Imagine Video 1.5 Lite #17 on its text-to-video leaderboards. The model sits on the quality-and-speed Pareto frontier, generating 10-second 1080p clips in a median of 60.5 seconds. Priced at $0.14 per second for 1080p output, it ranks ahead of Google's Veo 3.1 at approximately one-third of the cost.

Artificial AnalysisOct 8

Artificial Analysis Expands Cyber Index to Include Trusted-Access Frontier Models

Artificial Analysis expanded its Cyber Index to include trusted-access models, providing a broader view of frontier cyber defense capabilities. OpenAI’s GPT-6 Sol (Daybreak Blue, max) debuts at #1 with a 69% score and zero safety blocks. This model outperforms the public GPT-6 Sol (max) by 32 points at a cost of $1.77 per task.

Artificial AnalysisOct 8

Artificial Analysis Updates Harvey LAB-AA Benchmark with New Hallucination Gating

Artificial Analysis launched Harvey LAB-AA v1.1, adding a hallucination check to its legal agent benchmark. The new Hallucination-Gated All-Pass Rate credits tasks only when deliverables satisfy rubric criteria without material hallucinations. Grok 4.7 leads at 9.4%, while the evaluation reveals that over 60% of otherwise passing results across tested models contain material hallucinations.

Artificial AnalysisOct 8

Artificial Analysis Announces Intelligence Index v5 for Late October

Artificial Analysis will release Intelligence Index v5 in late October, marking a significant update to its frontier AI benchmarking. The new version incorporates Terminal-Bench Science and introduces a coding benchmark utilizing a private dataset. These additions aim to provide a more comprehensive evaluation of AI model capabilities.

Artificial AnalysisOct 8

Artificial Analysis Ranks Google Nano Banana 2.1 Fourth on Image Boards

Artificial Analysis ranks Google's new Nano Banana 2.1 model fourth on its Text-to-Image and Image Editing leaderboards. Built on Gemini 3.6 Flash, the model outperforms its predecessor across all 16 measured capabilities while costing $0.0336 per 1K image, half the price of Nano Banana 2. It is now available across Google's Gemini ecosystem and third-party providers.

Artificial AnalysisOct 8

Artificial Analysis Video Leaderboards Add Derived Model Labels and Filters

Artificial Analysis now labels derived video models with their base model's logo on its leaderboards. A new filter allows users to hide these post-trained models to rank base models exclusively. The update clarifies model lineage, as derived variants like Utopai X and MiniMax H3 Max often differ from their base model in pricing, speed, and availability.

Artificial AnalysisOct 8

Artificial Analysis Ranks Vidu Q4 Preview Third on Video Leaderboard

Artificial Analysis ranks Vidu Q4 Preview third on its AA-Video-I2V v1.0 leaderboard, a 16-place improvement over Vidu Q3 Pro. The model generates 3 to 16-second clips at up to 4K with native audio, supporting 15 image and 3 audio references. It maintains the $7.20 per minute price point of the previous generation while delivering higher quality.

Artificial AnalysisOct 8

Artificial Analysis Updates Music Arena Benchmarks with New Vocal and Instrumental Leaderboards

Artificial Analysis launched AA-Music-Vocal v1.1 and AA-Music-Instrumental v1.1, featuring 1,000 new prompts across 17 genres. Ratings are derived from over 37,000 blind human preference votes. Suno v6 holds the top two spots on both leaderboards, while MiniMax Music 3.0 ranks as the leading open-weights model.

Artificial AnalysisOct 8

Artificial Analysis Benchmarks Claude Haiku 5.5 Intelligence and Pricing

Artificial Analysis reports that Anthropic's Claude Haiku 5.5 scores 43 on its Intelligence Index, a 26-point increase over the prior release. The model introduces tiered pricing, effort settings, and a 1M token context window. While it demonstrates high performance on agentic knowledge work and terminal tasks, it shows high output token consumption at max effort settings.

Artificial AnalysisOct 6

Artificial Analysis Benchmarks Mistral Large 4 Preview Intelligence and Cyber Capabilities

Artificial Analysis reports that Mistral Large 4 Preview scores 38 on its Intelligence Index, making it the most intelligent model from outside the US and China. The 1T-parameter model achieves a 50 on the Cyber Index and supports text and image input. Mistral plans to release the model weights at the end of October.

Artificial AnalysisOct 6

Artificial Analysis Benchmarks OpenAI Web Search on Search Index

Artificial Analysis benchmarked OpenAI Web Search on its Search Index, where it debuted at 74. This places it as the 5th best provider, trailing Perplexity, Octen, Parallel, and Brave. While OpenAI’s tool is cheaper per task than most Search APIs at about $0.05, it performs best on factual benchmarks and trails on multi-hop reasoning tasks.

Artificial AnalysisOct 6

Artificial Analysis Ranks ElevenLabs Eleven v4 Turbo #1 in TTS

Artificial Analysis ranks ElevenLabs Eleven v4 Turbo first on its Provider Voice TTS Arena Leaderboard with an Elo of 1,334. The model generates speech at 96 characters per second, outperforming the flagship Eleven v4 while costing $40 per million characters—half the price of the previous version. It also achieves 90.1% accuracy on the firm's pronunciation robustness benchmark.

Artificial AnalysisOct 3

Artificial Analysis Ranks Ideogram 4.5 on Image Editing and T2I Leaderboards

Artificial Analysis ranks Ideogram 4.5 at #23 on its Image-Editing v2.0 leaderboard and #34 on its Text-to-Image v2.0 leaderboard. This marks the first time an Ideogram model has appeared on the editing board. At the high-quality level evaluated, the model costs $0.22 per image for edits and $0.10 per image for Text-to-Image generation.

Artificial AnalysisOct 2

Artificial Analysis Updates Coding Agent Index With Three New Top Models

Artificial Analysis updated its Coding Agent Index with three new top-performing agents. Claude Sonnet 5.5 in Claude Code leads with a score of 68 at $14.19 per task. Gemini 4 Argon scores 64 at $5.84, while GPT-6.1 Sol scores 63 at $1.04. The index composites performance across three software engineering benchmarks, highlighting distinct performance-to-cost trade-offs for each agent.

Artificial AnalysisOct 2

Artificial Analysis Ranks Alibaba Qwen-Image-2.1 as Top Open Weights Model

Artificial Analysis ranks Alibaba's Qwen-Image-2.1 as the leading open weights model on its text-to-image and image-editing leaderboards. The 7B-parameter model, which supports native 2K output and RGBA transparency, climbed to 18th overall on both boards. It is available on Hugging Face and ModelScope with day-0 support in ComfyUI, Diffusers, and other local inference tools.