Model Releases
New foundation models, benchmarks, and capability breakthroughs.
OpenAI's New Benchmark Measures the One Thing That Actually Matters: Can AI Do Your Job?
1 minutes agoOpenAI's New Benchmark Measures the One Thing That Actually Matters: Can AI Do Your Job?
GPT-5.5 scores 84.9% on GDPVal — OpenAI's new benchmark spanning 44 real occupations across 9 industries — and the trajectory of that number tells a more uncomfortable story than any academic test score.
Mistral Just Quietly Launched the Coding Agent That Opens Your PRs While You Sleep
1 minutes agoMistral Just Quietly Launched the Coding Agent That Opens Your PRs While You Sleep
Mistral's Medium 3.5 scores 77.6% on SWE-Bench Verified and introduces cloud-hosted remote coding agents that autonomously execute, test, and submit code changes as GitHub pull requests.
OpenAI Just Shattered the Voice AI Stack Into Three — and That's Exactly What the Industry Needed
1 minutes agoOpenAI Just Shattered the Voice AI Stack Into Three — and That's Exactly What the Industry Needed
OpenAI launches GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper on May 7, decomposing monolithic voice AI into composable infrastructure primitives with task-specific pricing.
xAI's Grok 4.3 Arrives at $1.25 Per Million Tokens — and Bundles a Voice Clone API That Changes Enterprise AI
21 hours agoxAI's Grok 4.3 Arrives at $1.25 Per Million Tokens — and Bundles a Voice Clone API That Changes Enterprise AI
xAI releases Grok 4.3 at $1.25/M tokens with 1M context window, agentic optimization, and a voice cloning API with anti-deepfake liveness verification.
Grok Imagine 1.0 Just Generated a Billion Videos a Month — and the World Hasn't Noticed the Real Shift Yet
1 days agoGrok Imagine 1.0 Just Generated a Billion Videos a Month — and the World Hasn't Noticed the Real Shift Yet
xAI launched Grok Imagine 1.0 in February 2026, hitting 1.245 billion monthly videos and establishing the largest synthetic media platform ever built.
Claude Opus 4.7 Quietly Raised the Bar—And the Industry Did Not Notice How High
1 days agoClaude Opus 4.7 Quietly Raised the Bar—And the Industry Did Not Notice How High
Anthropic's Opus 4.7 hits 87.6% on SWE-bench Verified, up from 80.8%, with 3x better image resolution and built-in safeguards against cybersecurity misuse.
Grok 5's 6 Trillion Parameters Are Not a Technical Spec — They're xAI's Political Statement About Who Wins the AI Race
1 days agoGrok 5's 6 Trillion Parameters Are Not a Technical Spec — They're xAI's Political Statement About Who Wins the AI Race
xAI's Grok 5 is set to debut in Q2 2026 as the largest AI model ever announced at 6 trillion parameters, trained on a gigawatt-scale cluster with exclusive Tesla fleet and X platform data.
Microsoft Just Declared Its Independence from OpenAI — and Its Three New Models Are the Proof
1 days agoMicrosoft Just Declared Its Independence from OpenAI — and Its Three New Models Are the Proof
Microsoft released MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 through Azure Foundry, outperforming OpenAI's Whisper on all 25 languages and entering the voice cloning and image generation markets.
Google's Budget AI Model Just Tripled Its Intelligence Score — And That Should Terrify Every Enterprise Paying Frontier Prices
1 days agoGoogle's Budget AI Model Just Tripled Its Intelligence Score — And That Should Terrify Every Enterprise Paying Frontier Prices
Gemini 3.1 Flash-Lite nearly triples its composite intelligence score while pricing at $0.25 per million tokens, challenging the economic logic of frontier AI for most enterprise workloads.
Moonshot AI's 1-Trillion-Parameter Open-Source Gamble Is the Most Disruptive Model Release Nobody Is Talking About
1 days agoMoonshot AI's 1-Trillion-Parameter Open-Source Gamble Is the Most Disruptive Model Release Nobody Is Talking About
Kimi K2.5 combines 1.04T parameters with a native 100-agent swarm under MIT license — benchmark results rival closed frontier models at a fraction of API costs.
DeepSeek V4 Pro Doesn't Just Match Claude — It Does It at 86% Off the Sticker Price
2 days agoDeepSeek V4 Pro Doesn't Just Match Claude — It Does It at 86% Off the Sticker Price
DeepSeek's April 24 release: a 1.6T-parameter MIT-licensed model scoring 80.6% on SWE-bench Verified at $3.48 per million output tokens versus Claude's $25.
China's GLM-5.1 Just Topped Every US AI Model at Coding — and It Didn't Use a Single Nvidia Chip
2 days agoChina's GLM-5.1 Just Topped Every US AI Model at Coding — and It Didn't Use a Single Nvidia Chip
Z.ai's GLM-5.1 scored 58.4% on SWE-Bench Pro in April 2026, beating GPT-5.4 and Claude Opus 4.6 to become the first open-weight model to top the global coding benchmark, trained entirely on Huawei hardware.
The MIT License Is Now a Weapon: Z.ai's GLM-5.1 Just Beat GPT-5.4 and Anyone Can Use It
2 days agoThe MIT License Is Now a Weapon: Z.ai's GLM-5.1 Just Beat GPT-5.4 and Anyone Can Use It
Z.ai released GLM-5.1 on April 7, 2026: a 744B open-source model under MIT license that topped SWE-Bench Pro, beating GPT-5.4 and Claude Opus 4.6.
DeepSeek V4 Flash Just Killed the Inference Pricing Premium — and Nobody Noticed
3 days agoDeepSeek V4 Flash Just Killed the Inference Pricing Premium — and Nobody Noticed
DeepSeek V4 Flash — a 284B MoE model activating just 13B parameters per token at $0.14 per million inputs — has permanently reset the economics of AI inference and exposed the fragility of mid-tier pricing models.
Google Just Made Open-Source AI 3x Faster Without Losing a Single Token of Quality — and That Changes the Inference Economics
4 days agoGoogle Just Made Open-Source AI 3x Faster Without Losing a Single Token of Quality — and That Changes the Inference Economics
Google released Multi-Token Prediction drafters for Gemma 4 on May 6, 2026, delivering up to 3x faster inference with zero quality degradation under the Apache 2.0 open-source license.
Gemini 3.1 Ultra's 2-Million-Token Brain Isn't Google's Real Move — The Strategy Behind It Is
4 days agoGemini 3.1 Ultra's 2-Million-Token Brain Isn't Google's Real Move — The Strategy Behind It Is
Google's Gemini 3.1 Ultra scored 94.3% on GPQA Diamond and processes 1,500+ pages in a single session — but the real story is how Google just repositioned itself at the center of the multimodal AI race.
ChatGPT Just Got a New Brain — And What It's Quietly Admitting About the Hallucination Problem
4 days agoChatGPT Just Got a New Brain — And What It's Quietly Admitting About the Hallucination Problem
OpenAI replaced ChatGPT's default model with GPT-5.5 Instant on May 5, claiming 52.5% fewer hallucinations and 30% shorter responses in high-stakes domains.
The Math Problem Defining Every AI Model Since 2017 May Finally Have a Solution — And a $29M Bet on It
4 days agoThe Math Problem Defining Every AI Model Since 2017 May Finally Have a Solution — And a $29M Bet on It
Miami startup Subquadratic raises $29M seed and launches SubQ, claiming a 12M-token context window with 1,000x attention compute reduction via subquadratic sparse attention.
OpenAI's GPT-5.5 Instant Slashes Hallucinations by Half — and the Real Story Is What That Changes for Enterprise AI
4 days agoOpenAI's GPT-5.5 Instant Slashes Hallucinations by Half — and the Real Story Is What That Changes for Enterprise AI
OpenAI replaced ChatGPT's default model with GPT-5.5 Instant on May 5, 2026, reporting 52.5% fewer hallucinations on medical, legal, and financial queries.
DeepSeek V4 on Huawei Chips Is Not a Benchmark Story. It Is a Supply Chain Story That Changes Everything.
5 days agoDeepSeek V4 on Huawei Chips Is Not a Benchmark Story. It Is a Supply Chain Story That Changes Everything.
DeepSeek V4-Pro with 1.6T parameters trained on Huawei Ascend chips at $3.48/million tokens challenges the core logic of US AI chip export controls.
The AI That Rewrote Itself 100 Times — And What MiniMax M2.7 Means for Labs Betting on Human Supervision
5 days agoThe AI That Rewrote Itself 100 Times — And What MiniMax M2.7 Means for Labs Betting on Human Supervision
MiniMax M2.7 is a self-evolving 230B model that matches Claude Opus 4.6 on SWE-Pro at 50x lower cost after autonomously running 100+ optimization rounds.
Meta Just Abandoned Its Own Religion: Muse Spark Is Proprietary and the Open Source AI Story Will Never Be the Same
May 5, 2026Meta Just Abandoned Its Own Religion: Muse Spark Is Proprietary and the Open Source AI Story Will Never Be the Same
Meta's first model from Alexandr Wang's Superintelligence Labs ranks 4th globally, abandoning Llama's open source tradition with a proprietary license and a HealthBench score triple any competitor's.
NVIDIA's Open Model Delivers 9x the Throughput of Every Competitor — For Free
May 4, 2026NVIDIA's Open Model Delivers 9x the Throughput of Every Competitor — For Free
NVIDIA's Nemotron 3 Nano Omni processes video, audio, and documents simultaneously in 25GB of RAM, outperforming Qwen3-Omni ninefold on throughput — and the benchmarks reveal a hardware strategy hiding inside an open-source gift.
AlphaEvolve: Google's Self-Improving AI Just Cracked a Math Problem That Stumped Humans for 55 Years
May 4, 2026AlphaEvolve: Google's Self-Improving AI Just Cracked a Math Problem That Stumped Humans for 55 Years
Google DeepMind's AlphaEvolve pairs Gemini with evolutionary algorithms to crack open math problems, optimize data centers, and speed up AI training autonomously.
Alibaba's Qwen 3.6-Plus Just Made the Closed-Weights Arms Race Look Obsolete
May 3, 2026Alibaba's Qwen 3.6-Plus Just Made the Closed-Weights Arms Race Look Obsolete
Alibaba's Qwen 3.6-Plus delivers a 1M-token context window, 3x Claude Opus 4.6 inference speed, and top SWE-bench Pro scores — now on Fireworks AI with production SLAs for enterprise inference.
GPT-5.5, Claude Mythos, and Gemini 3.1 Pro All Won Spring 2026 — What That Means for Everyone Building on AI
May 3, 2026GPT-5.5, Claude Mythos, and Gemini 3.1 Pro All Won Spring 2026 — What That Means for Everyone Building on AI
GPT-5.5 leads agents, Claude Mythos leads coding at 93.9% SWE-Bench, Gemini 3.1 Pro leads reasoning—spring 2026 ended the era of a single best AI model.
OpenAI Built a Cyberweapon and Called It a Shield — The Uncomfortable Math of GPT-5.5-Cyber
May 3, 2026OpenAI Built a Cyberweapon and Called It a Shield — The Uncomfortable Math of GPT-5.5-Cyber
OpenAI's GPT-5.5-Cyber scores 81.8% on CyberGym and is being deployed exclusively to government and critical infrastructure defenders via its Trusted Access for Cyber program.
Mistral's 128B Gamble: One Model to Replace Three — and Why Europe's AI Challenger Is Playing a Different Game
May 3, 2026Mistral's 128B Gamble: One Model to Replace Three — and Why Europe's AI Challenger Is Playing a Different Game
Mistral Medium 3.5 consolidates three prior models into a single 128B open-weight architecture scoring 77.6% on SWE-Bench Verified, released April 29, 2026.
Gemini Robotics-ER 1.6 Can Read a Lab Instrument. That Changes Everything About the Robots We Are Building.
May 3, 2026Gemini Robotics-ER 1.6 Can Read a Lab Instrument. That Changes Everything About the Robots We Are Building.
Google DeepMind's Gemini Robotics-ER 1.6 adds instrument reading, self-failure detection, and enhanced spatial reasoning to physical AI — launching April 15, 2026 with cross-embodiment transfer.
America Paid $200 for This Coding AI. China Built a Better One for $3.
Zhipu AI GLM-4.7 posts 73.8% on SWE-bench and 84.9 on LiveCodeBench V6, beating Claude Sonnet 4.5, at $3/month — up to 66x cheaper than US coding AI tools.
Moonshot AI's Kimi K2.6 Just Beat GPT-5.4 at Coding — With 300 AI Agents Running in Parallel
May 2, 2026Moonshot AI's Kimi K2.6 Just Beat GPT-5.4 at Coding — With 300 AI Agents Running in Parallel
Moonshot AI's Kimi K2.6 beats GPT-5.4 on SWE-Bench Pro with a 300-agent swarm, 256K context, and Modified MIT license — released open-weight April 20, 2026.
GLM-5.1 by Z.ai Tops SWE-Bench Pro, Beating GPT-5.4 and Claude
Z.ai's GLM-5.1 tops SWE-Bench Pro with 58.4, surpassing GPT-5.4 and Claude Opus 4.6 — the first open-weight model to beat closed frontier AI on coding.
Google Gemma 4 Apache 2.0: Why the License Matters More Than the Benchmark
Google released Gemma 4 under Apache 2.0 on April 2, 2026 — four model sizes from 2B to 31B, with full commercial freedom, targeting Meta Llama dominance.
Meta Llama 4 Scout 10M Context vs RAG Enterprise Strategy
Meta releases open-weight Llama 4 Scout (10M-token context) and Maverick (beats GPT-4o), challenging RAG-based enterprise AI architecture
Alibaba Qwen 3.5 Rewrites the AI Cost Equation
Alibaba Qwen 3.5: 397B open-weight MoE, 201 languages, 1M-token context, Apache 2.0, outperforms larger models at fraction of inference cost
Google I/O 2026 Preview: Gemini 2.5 Ultra, AI-Only Search Mode, and the End of the Ten Blue Links
Apr 28, 2026Google I/O 2026 Preview: Gemini 2.5 Ultra, AI-Only Search Mode, and the End of the Ten Blue Links
Google is expected to unveil its most aggressive AI-first product lineup ever at I/O 2026 -- fundamentally restructuring how a billion users find information.
Google Bets Its Cloud Future on Autonomous AI Agents
At Cloud Next 2026, Sundar Pichai unveiled 8th-gen TPUs and a full agentic platform designed to rewire enterprise automation.
OpenAI Releases GPT-5.5: Smarter Reasoning, Less Instruction Required
The latest flagship model arrives six weeks after GPT-5.4, scoring 82.7% on Terminal-Bench 2.0 and outpacing Claude Opus 4.7 and Gemini 3.1 Pro on agentic task benchmarks.
The AI Model Race Enters Its Most Competitive Era Yet
From Google's 8th-gen TPUs to open-weight models closing the frontier gap, 2026 is reshaping who controls AI.
The Great Model Race of 2026 Is Reshaping AI's Power Map
From Google's trillion-parameter Gemini to Moonshot AI's open-source Kimi K2.6, a new wave of releases is redrawing competitive lines.
The April 2026 Model Surge That Redrew the AI Frontier
From Claude Opus 4.7 to GPT-5 Turbo, a single month reshaped what foundation models can do.
The 2026 Model Surge Is Rewriting the Rules of AI Competition
With 291 releases tracked by April and benchmark gaps narrowing fast, the AI model race has entered a new, more brutal phase.
The Great Model Flood: AI's Release Velocity Hits an Inflection Point
With 291 models tracked in early 2026 alone, the industry is reshaping competition, pricing, and the very meaning of frontier AI.
The AI Model Wars of 2026 Are Rewriting the Frontier
From OpenAI's o3 to NVIDIA's Physical AI platform, a relentless wave of releases is compressing the performance gap and reshaping enterprise computing.