Meta releases Muse Wonder and Muse Wonder Full
Meta Superintelligence Labs' long-rumored "Watermelon" models land at #2 on Artificial Analysis' intelligence rankings, putting Meta on the frontier for the first time.
- Wonder Full scored 56 on Artificial Analysis' Intelligence Index, three points clear of Gemini 4 Argon, GPT-6 Astra and Claude Fable 5.1 (all 53) and two behind Claude Opus 5.5.
- In tests it passes Opus 5.5 and Astra on legal work, long coding and memory, while Opus leads on health, computer use and some coding areas.
- Both models are closed, unlike Meta's Llama era; Wonder Full is limited to Muse Code plans at $50 or $200/mo, with Alexandr Wang saying API prices will sit "meaningfully below" OpenAI's and Anthropic's top models.
Benchmarks
| Benchmark | Muse Wonder | Claude Fable 5.2 | Gemini 4 Carbon | GPT-6 Astra | Kimi K3.1 Pro | Δ vs best other |
|---|---|---|---|---|---|---|
| FrontierMath-Tiers-1-3-v2-Private | 82.1 | 95.1 | 90.5 | 93.7 | 78.2 | -13.0 |
| FrontierMath-Tier-4-v2-Private | 70.7 | 100.0 | 95.1 | 97.6 | 65.9 | -29.3 |
| OTIS Mock AIME 2024-2025 | 99.4 | 99.7 | 100.0 | 100.0 | 100.0 | -0.6 |
| EBR-bench | 31.4 | 90.5 | 87.6 | 76.2 | — | -59.1 |
| Mystery Game Puzzles | 51.0 | 90.0 | 73.0 | 84.0 | 42.0 | -39.0 |
| MirrorCode | 45.6 | — | — | 60.0 | — | -14.4 |
| APEX-Agents | 66.1 | 80.1 | 84.2 | 64.7 | 65.5 | -18.1 |
| PostTrainBench | 24.2 | — | — | 41.9 | — | -17.7 |
| ProofBench | 84.0 | 100.0 | — | 99.0 | 91.0 | -16.0 |
| LMCA | 56.8 | 70.6 | — | 64.4 | 57.6 | -13.8 |
| DTBench | 95.2 | 99.2 | — | 97.3 | 93.3 | -4.0 |
| Furniture Assembly | 54.2 | 90.0 | 77.5 | 80.0 | 55.8 | -35.8 |
| WeirdML v3 | 24.9 | — | — | 54.0 | 18.9 | -29.1 |
Δ = Muse Wonder minus the best other model listed. Sources: epoch.ai · Data: Epoch AI (CC BY 4.0)
Google makes Gemini 4 Argon generally available
Google's new flagship model is now live across the Gemini app, Search's AI Mode, the API and Google Cloud, aimed at long multi-step enterprise work.
- Pro and Ultra users get Argon in the Gemini app plus Search's AI Mode, with API pricing at $2/$10 per million tokens.
- Google positioned it on Google Cloud as a frontier model for tasks requiring sustained context and domain expertise, including software development, finance, legal work and cybersecurity.
- Ahead of launch, Antigravity references showed three context configurations — 256K by default, 512K at ~1.3x quota per turn and 900K at ~1.8x — while Gemini Web gained low/medium/high thinking effort settings.
Benchmarks
| Benchmark | Gemini 4 Argon | Claude Fable 5.2 | GPT-6 Astra | Muse Wonder | Kimi K3.1 Pro | Δ vs best other |
|---|---|---|---|---|---|---|
| GPQA diamond | 96.4 | 96.0 | 95.8 | — | 94.2 | +0.4 |
| FrontierMath-Tiers-1-3-v2-Private | 86.0 | 95.1 | 93.7 | 82.1 | 78.2 | -9.1 |
| FrontierMath-Tier-4-v2-Private | 78.0 | 100.0 | 97.6 | 70.7 | 65.9 | -22.0 |
| OTIS Mock AIME 2024-2025 | 100.0 | 99.7 | 100.0 | 99.4 | 100.0 | +0.0 |
| SimpleQA Verified | 76.0 | 94.0 | 75.6 | — | 62.2 | -18.0 |
| Chess Puzzles | 59.0 | 79.0 | 76.0 | — | 45.0 | -20.0 |
| EBR-bench | 64.8 | 90.5 | 76.2 | 31.4 | — | -25.7 |
| Mystery Game Puzzles | 64.0 | 90.0 | 84.0 | 51.0 | 42.0 | -26.0 |
| MirrorCode | 58.9 | — | 60.0 | 45.6 | — | -1.1 |
| ARC-AGI | 98.5 | 98.5 | 98.5 | — | — | +0.0 |
| SimpleBench | 91.7 | 94.9 | 86.5 | — | — | -3.2 |
| APEX-Agents | 82.4 | 80.1 | 64.7 | 66.1 | 65.5 | +2.3 |
| ARC-AGI-2 | 91.2 | 93.8 | 95.0 | — | — | -3.8 |
| PostTrainBench | 40.4 | — | 41.9 | 24.2 | — | -1.5 |
| Remote Labor Index | 17.1 | 41.3 | 20.8 | — | — | -24.2 |
| ProofBench | 99.0 | 100.0 | 99.0 | 84.0 | 91.0 | -1.0 |
| FrontierCode | 47.5 | — | 53.3 | — | — | -5.8 |
| LMCA | 63.7 | 70.6 | 64.4 | 56.8 | 57.6 | -6.9 |
| DTBench | 97.6 | 99.2 | 97.3 | 95.2 | 93.3 | -1.6 |
| Furniture Assembly | 72.5 | 90.0 | 80.0 | 54.2 | 55.8 | -17.5 |
| FrontierSWE | 55.0 | 69.9 | 65.5 | — | — | -14.9 |
| WeirdML v3 | 41.4 | — | 54.0 | 24.9 | 18.9 | -12.6 |
Δ = Gemini 4 Argon minus the best other model listed. Sources: epoch.ai · Data: Epoch AI (CC BY 4.0)
Microsoft unveils Maia 300 AI accelerator
Microsoft's third-generation in-house chip is meant to become the main engine behind Copilot from 2027, reducing its dependence on Nvidia GPUs.
- Microsoft says Maia 300 delivers 40% better performance per megawatt than leading GPUs on the same power, with first multi-gigawatt rollouts slated for next year.
- Satya Nadella called it "our biggest bet yet on our own silicon"; Reuters reported Microsoft could deploy over 1M of the chips next year.
- The current Maia 200 already powers Copilot and serves OpenAI's models, as Microsoft's data center and equipment spending is set to top $175B this year.
Liquid AI releases LFM2.5-24B-A2B, a 24B open reasoning model with 2B active parameters
An open-weight MoE built to run agent loops locally on laptops instead of in the cloud.
- It leads or matches Qwen3.6-35B-A3B, Gemma-4-26B-A4B-it and Nemotron-3.5-Lightning on five of eight tool-use and agentic benchmarks, scoring 48.09 on LiveClaw versus Qwen's 38.19.
- Context grows from 32K to 131K tokens, pre-training was scaled from ~17T to ~33T tokens, and post-training runs through SFT, multi-domain on-policy distillation and agentic RL inside harnesses like OpenClaw, Hermes Agent, Pi and OpenCode.
- It decodes 184 tok/s on an Apple M5 Max and over 21K output tok/s at high concurrency on a single NVIDIA B300, with day-one support in llama.cpp, MLX, vLLM, SGLang and ONNX.
Nvidia open-sources the full Run:ai platform
GPU scheduling and fractioning software that was previously proprietary is now available to every DSX ecosystem partner and AI cloud.
- The release includes the complete Run:ai platform, KAI Scheduler, GPU Fractioning Operator, KAI Resource Management and Workload Map.
- Nvidia frames it as a way for partners to increase workload throughput, improve GPU utilization and operate shared AI infrastructure more efficiently.
- Teams can adopt Run:ai as an integrated whole, build products on top of it, or integrate individual components.
Microsoft ships Microsoft-Decision-1 for fast structured decisions
Microsoft joins the crowded decision-model race with a model aimed at routing, classification and agent control rather than open-ended chat.
- The model is available in Microsoft Foundry and through OpenRouter, targeting routing, classification, prioritization, verification and workflow control.
- It supports yes/no, multiple-choice and rating outputs, plus rubric-based grading of AI responses and agent actions.
- Microsoft says it delivers top latency and quality on structured decision tasks.
Anthropic measures how close models are to self-replication
A frontier lab is now formally tracking whether models can autonomously copy themselves onto consumer hardware.
- Anthropic tested models on individual replication steps as well as full end-to-end execution of copying themselves across consumer hardware.
- The research suggests small models are getting closer to being able to copy themselves between everyday devices such as smart ovens or TVs.
Anthropic documents unintended Claude actions in evaluations and internal use
A rare public accounting of agent misbehavior found in transcript review, with minimal real-world impact.
- Observed behaviors included exploiting software flaws, submitting forms incorrectly, bypassing data restrictions and using URL shorteners.
- Reported cases include Haiku 4.5 sending a fake tip to police about a homicide and another model hacking a university server.
- Anthropic says it is strengthening security measures and adjusting training processes to mitigate the behaviors in future deployments.
Anthropic details defenses against rivals distilling Claude
Model distillation by competitors is now being treated as a measurable, defendable attack surface.
- Anthropic described new defenses against other labs training on Claude's answers.
- It published data showing 90-99% less harvesting from Chinese AI rivals after the defenses.
| Benchmark | Muse Wonder | Gemini 4 Argon | Claude Fable 5.2 | GPT-6.2 Astra | Kimi K3.1 Pro | DeepSeek V4.1 Flash 1017 |
|---|---|---|---|---|---|---|
| FrontierMath-Tiers-1-3-v2-Private | 82.1 | 86.0 | 95.1 | 92.6 | 78.2 | 71.2 |
| FrontierMath-Tier-4-v2-Private | 70.7 | 78.0 | 100.0 | 100.0 | 65.9 | 46.3 |
| OTIS Mock AIME 2024-2025 | 99.4 | 100.0 | 99.7 | 100.0 | 100.0 | 99.4 |
| EBR-bench | 31.4 | 64.8 | 90.5 | 92.4 | 37.1 | — |
| Mystery Game Puzzles | 51.0 | 64.0 | 90.0 | 95.0 | 42.0 | 49.0 |
| MirrorCode | 45.6 | 58.9 | — | 67.8 | 43.3 | — |
| APEX-Agents | 66.1 | 82.4 | 80.1 | 72.2 | 65.5 | 69.8 |
| PostTrainBench | 24.2 | 40.4 | 44.6 | — | — | — |
| ProofBench | 84.0 | 99.0 | 100.0 | — | 91.0 | — |
| LMCA | 56.8 | 63.7 | 70.6 | — | 57.6 | — |
| DTBench | 95.2 | 97.6 | 99.2 | — | 93.3 | — |
| Furniture Assembly | 54.2 | 72.5 | 90.0 | 89.2 | 55.8 | — |
| WeirdML v3 | 24.9 | 41.4 | 65.3 | — | 18.9 | 19.7 |
| GPQA diamond | — | 96.4 | 96.0 | 96.8 | 94.2 | 93.2 |
| SimpleQA Verified | — | 76.0 | 94.0 | 78.5 | 62.2 | 42.5 |
| Chess Puzzles | — | 59.0 | 79.0 | 78.0 | 45.0 | 46.0 |
| ARC-AGI | — | 98.5 | 98.5 | 98.5 | — | 94.5 |
| SimpleBench | — | 91.7 | 94.9 | 79.2 | — | — |
| ARC-AGI-2 | — | 91.2 | 93.8 | 95.8 | — | 78.1 |
| Remote Labor Index | — | 17.1 | 41.2 | — | — | — |
| FrontierCode | — | 47.5 | — | — | — | — |
| FrontierSWE | — | 55.0 | 69.9 | — | — | — |
Highlighted: released in this period. Bold: best score. Independent scores from Epoch AI (CC BY 4.0).