unbenchmark.com is now in beta and receiving its first reviews. if you are tired of nonsense evals and do not trust AI models benchmarks, give unbenchmark a try. we are an open platform for user reviews of AI models. 100% free, self-funded and independent.

Anthropic IPO file is out, here are my 9 highlights: 👉 Top 5 customers represent ~25% of revenue 👉 Revenue: $16.3B in H1 2026, up from $1.2B in H1 2025 👉 Revenue grew 1,368% YoY while compute costs grew 300% 👉 Net Revenue Retention: ~900% 👉 Over 90% of the contributions to Anthropic's codebase in September 2026 were written by Claude 👉 Revenue per employee: $6.6M, up from $1.4M a year earlier 👉 Every company in the Fortune 10 spends $100K+ a year with Anthropic 👉 ~250 customers pay $10M+ a year 👉 Approximately 5,300 employees worldwide. Are you bullish or bearish on Anthropic's IPO?

Decision Models Launched in 2026: 👉 Jev 1.13 (Typesafe): Sep 15 👉 Solar Decide (Upstage): Sep 28 👉 D1 (Liquid AI): Sep 29 👉 Mercury Decide (Inception): Sep 30 👉 Clef (Cloudflare): Oct 1 👉 Clef Flash (Cloudflare): Oct 1 👉 Decider V1 27B (Perplexity): Oct 1 👉 Decider V1.1 27B (Perplexity): Oct 7 👉 Solar Decide Flash (Upstage): Oct 7 👉 Clef Omni (Cloudflare): Oct 9 👉 Microsoft-Decision-1 (Microsoft): Oct 9 👉 Jev 1.14 (Typesafe): Oct 14 👉 d1-ultra (Liquid AI): Oct 15 👉 Decider Flash V1.2 (Perplexity): Oct 16 👉 Decider V1.2 (Perplexity): Oct 16 👉 Elephant 1 (Baseten): Oct 21 👉 Mammoth 1 (Baseten): Oct 21 Data from @unbenchmark

Anthropic launched 11 models in 2026. Here's the breakdown: 👉 Opus 4.6 - Feb 5 👉 Sonnet 4.6 - Feb 17 👉 Opus 4.7 - Apr 16 👉 Opus 4.8 - May 28 👉 Fable 5 - Jun 9 👉 Sonnet 5 - Jun 30 👉 Opus 5 - Jul 24 👉 Fable 5.1 - Sep 1 👉 Opus 5.5 - Sep 22 👉 Sonnet 5.5 - Sep 28 👉 Haiku 5.5 - Oct 7 6 in the first half 5 in the second half (until today) Data from @unbenchmark

Do you think Kimi K3.1 is benchmaxxed?

Price Wars are here. And this is just the beginning. Replit jumped ship to OpenAI - a deal worth $100M+/year with 30%+ off - right after Anthropic pulled its discount.

do you trust benchmarks?

CELEBRATE WITH ME! @unbenchmark JUST CROSSED 200+ REVIEWS Thank you all. Let's build a more transparent AI world. The one that do not rely solely on benchmarks. If you are not with us yet, you can sign up and start reviewing models for free, sharing their scores based on YOUR EXPERIENCE, not nonsense evals or benchmarks. Join us: unbenchmark.com

So apparently Moonshot was paying pages and influencers to hype up the Kimi K3.1 launch and telling them not to disclose that the posts were paid promotions? Hmmmmm

Open-Weight AI Models Released in 2026 (So Far) 👉 FLUX.2 Klein 4B · Black Forest Labs 🇩🇪 · Jan 15 👉 GLM 4.7 Flash · Z.ai 🇨🇳 · Jan 19 👉 LFM2.5-1.2B-Thinking · Liquid AI 🇺🇸 · Jan 20 👉 LFM2.5-1.2B-Instruct · Liquid AI 🇺🇸 · Jan 20 👉 Kimi K2.5 · Moonshot AI 🇨🇳 · Jan 27 👉 Step 3.5 Flash · StepFun 🇨🇳 · Jan 29 👉 Qwen3 ASR 1.7B · Alibaba 🇨🇳 · Jan 29 👉 Qwen3 ASR 0.6B · Alibaba 🇨🇳 · Jan 29 👉 Qwen3 Coder Next · Alibaba 🇨🇳 · Feb 4 👉 GLM 5 · Z.ai 🇨🇳 · Feb 11 👉 MiniMax M2.5 · MiniMax 🇨🇳 · Feb 12 👉 Qwen3.5 397B A17B · Alibaba 🇨🇳 · Feb 16 👉 Qwen3.5-35B-A3B · Alibaba 🇨🇳 · Feb 24 👉 Qwen3.5-27B · Alibaba 🇨🇳 · Feb 24 👉 Qwen3.5-122B-A10B · Alibaba 🇨🇳 · Feb 24 👉 LFM2-24B-A2B · Liquid AI 🇺🇸 · Feb 25 👉 Qwen3.5-9B · Alibaba 🇨🇳 · Mar 2 👉 Reka Edge · Reka AI 🇺🇸 · Mar 11 👉 Nemotron 3 Super · NVIDIA 🇺🇸 · Mar 11 👉 Mistral Small 4 · Mistral AI 🇫🇷 · Mar 16 👉 MiniMax M2.7 · MiniMax 🇨🇳 · Mar 18 👉 Gemma 4 26B A4B · Google 🇺🇸 · Mar 31 👉 Trinity Large Thinking · Arcee AI 🇺🇸 · Apr 1 👉 Gemma 4 31B · Google 🇺🇸 · Apr 2 👉 GLM 5.1 · Z.ai 🇨🇳 · Apr 7 👉 Qwen3.6 35B A3B · Alibaba 🇨🇳 · Apr 16 👉 Kimi K2.6 · Moonshot AI 🇨🇳 · Apr 20 👉 MiMo-V2.5-Pro · Xiaomi 🇨🇳 · Apr 22 👉 MiMo-V2.5 · Xiaomi 🇨🇳 · Apr 22 👉 Qwen3.6 27B · Alibaba 🇨🇳 · Apr 22 👉 DeepSeek V4 Flash · DeepSeek 🇨🇳 · Apr 23 👉 DeepSeek V4 Pro · DeepSeek 🇨🇳 · Apr 24 👉 Laguna XS.2 · Poolside 🇺🇸 · Apr 28 👉 Laguna M.1 · Poolside 🇺🇸 · Apr 28 👉 Nemotron 3 Nano Omni · NVIDIA 🇺🇸 · Apr 28 👉 Granite 4.1 8B · IBM 🇺🇸 · Apr 29 👉 Hy-MT2-7B · Tencent 🇨🇳 · May 21 👉 Hy-MT2-1.8B · Tencent 🇨🇳 · May 21 👉 Hy-MT2-30B-A3B · Tencent 🇨🇳 · May 21 👉 Step 3.7 Flash · StepFun 🇨🇳 · May 29 👉 MiniMax M3 · MiniMax 🇨🇳 · Jun 1 👉 Nex-N2-Mini · Nex AGI 🇨🇳 · Jun 2 👉 Nex-N2-Pro · Nex AGI 🇨🇳 · Jun 2 👉 Nemotron 3.5 Content Safety · NVIDIA 🇺🇸 · Jun 4 👉 Nemotron 3 Ultra · NVIDIA 🇺🇸 · Jun 4 👉 North Mini Code · Cohere 🇨🇦 · Jun 9 👉 Kimi K2.7 Code · Moonshot AI 🇨🇳 · Jun 12 👉 GLM 5.2 · Z.ai 🇨🇳 · Jun 13 👉 LFM2.5-Embedding-350M · Liquid AI 🇺🇸 · Jun 18 👉 LongCat 2.0 · Meituan 🇨🇳 · Jun 30 👉 Laguna XS 2.1 · Poolside 🇺🇸 · Jul 2 👉 Hy3 · Tencent 🇨🇳 · Jul 6 👉 Inkling · Thinking Machines Lab 🇺🇸 · Jul 15 👉 Kimi K3 · Moonshot AI 🇨🇳 · Jul 16 👉 Qwen3.8 2.4T A95B · Alibaba 🇨🇳 · Jul 19 👉 Laguna S 2.1 · Poolside 🇺🇸 · Jul 21 👉 Ling 3.0 Flash · inclusionAI 🇨🇳 · Jul 23 👉 Inkling Small · Thinking Machines Lab 🇺🇸 · Jul 30 👉 LFM2.5-2.6B · Liquid AI 🇺🇸 · Aug 4 👉 Ling 3.0 Tiny · inclusionAI 🇨🇳 · Aug 6 👉 Muse Glimmer 30B · Meta 🇺🇸 · Aug 10 👉 Nemotron 3.5 Lightning · NVIDIA 🇺🇸 · Aug 11 👉 GLM 5.3 · Z.ai 🇨🇳 · Aug 14 👉 Qwen3.8 27B · Alibaba 🇨🇳 · Aug 14 👉 DeepSeek V4 Flash Vision Exp · DeepSeek 🇨🇳 · Aug 21 👉 Granite 4.2 8B · IBM 🇺🇸 · Aug 25 👉 GLM 5.3 Flash · Z.ai 🇨🇳 · Aug 26 👉 Qwen3.8 Flash · Alibaba 🇨🇳 · Aug 26 👉 DeepSeek V4.1 Flash · DeepSeek 🇨🇳 · Sep 10 👉 MiMo-V2.6-Pro · Xiaomi 🇨🇳 · Sep 21 👉 MiMo-V2.6-Flash · Xiaomi 🇨🇳 · Sep 21 👉 Clef Flash · Cloudflare 🇺🇸 · Oct 1 👉 Clef · Cloudflare 🇺🇸 · Oct 1 👉 Beam 1.0 501B A23B · Reflection 🇺🇸 · Oct 5 👉 Clef Omni · Cloudflare 🇺🇸 · Oct 9 👉 LFM2.5-24B-A2B · Liquid AI 🇺🇸 · Oct 13 👉 Muse Spark Open · Meta 🇺🇸 · Oct 15 👉 Mammoth 1 · Baseten 🇺🇸 · Oct 21 👉 Elephant 1 · Baseten 🇺🇸 · Oct 21 👉 Flint · Recursive 🇺🇸 · Oct 21 Data from @unbenchmark 🇨🇳 47 · 🇺🇸 29 · 🇩🇪 1 · 🇫🇷 1 · 🇨🇦 1

benchmark wins are not always user wins

This was 6 months ago btw

impressive benchmark still wouldn't use it

Agents will bring an era of a loooot of confusion before things stabilize

Things we got before GTA VI: 👉 AGI 👉 Flying cards 👉 AI girlfriends 👉 Robots doing backflips 👉 People getting fired by a chatbot 👉 Coding without knowing how to code what else?

here's my eval: open model use model decide if model sucks