📂 LLM & Models
Comparisons, benchmarks and practical guides on large language models: GPT, Claude, Gemini, Llama and free alternatives.
GPT-6.2 Astra and Sol: OpenAI crosses its own self-improvement threshold, at one-third the cost per task
**Claude Mythos ports the AMD MI355X kernels in 3 days instead of a year — near-zero human intervention**
Claude Mythos ported the AMD MI355X kernels in 3 days instead of a year, with almost no human intervention. A major breakthrough announced by Anthropic.
Qwen4 Max: Alibaba's new flagship lands on the day open source accounts for 10% of servers
Alibaba launches Qwen4 Max, its new flagship AI model, as Chinese open source now accounts for 10% of servers. An analysis of ironic timing.
Google launches Gemini Decision 1: up to 1,000 scored questions per call, $0.10/M tokens for input, and zero cost for output
Google launches Gemini Decision 1, an AI model that generates no text: 1,000 scored questions per call, $0.10/M input tokens, $0 output.
Meituan open-sources LongCat-2.5: a 1600B-parameter multimodal MoE (48B active, 1M context) under MIT license
Meituan open-source LongCat-2.5: a 1.6T-parameter multimodal MoE, 48B active, 1M context, MIT licensed. Breakdown.
Anthropic launches Fable 5.2 and Mythos 5.2: 30% faster, 20% cheaper, a week ahead of its $100B IPO
Anthropic launches Fable 5.2 and Mythos 5.2: models 30% faster, 20% cheaper, one week before its $100B IPO. Analysis of the timing.
Thinking Machines unveils Inkling 2 Small: Apache 2.0 weights next week
Thinking Machines unveils Inkling 2 Small, a 594B-parameter MoE model. Its Apache 2.0 weights will be released next week.
Best Local LLMs (November 2026)
Discover the best local LLMs of November 2026: our ranking of open weights models to run on your own machine.
Best Free LLMs (November 2026)
Discover the best free LLMs in November 2026: Gemini 3.1 Pro, GPT-5.5, Claude Opus 4.7... the complete ranking of free AIs that are truly worth *(143 characters — tool names kept as-is, source text appears truncated mid-sentence)*
Kimi K3.1 Pro: Moonshot opens up a 2.8 trillion-parameter MoE with a one million token context window
Kimi K3.1 Pro: Moonshot AI releases the weights of its 2.8 trillion parameter MoE with a million-token context on Hugging Face.
Baidu releases ERNIE 5.5 Flash Base: the flash model race extends to China
Baidu releases ERNIE 5.5 Flash Base on Hugging Face: a 280B-parameter MoE model with 14B active parameters, extending the flash model race in
NVIDIA Nemotron 3.5 Transcribe: the open-weight transcription model for 40 languages, and Granary v2 (463,000 hours of speech)
NVIDIA Nemotron 3.5 Super & Lightning: open-weights hybrid Mamba-Transformer MoEs tailored for agentic AI
NVIDIA Nemotron 3.5 Super & Lightning: open-weights hybrid Mamba-Transformer MoE models optimized for agentic AI and long contexts.
Gemini 4 Carbon: Google retakes first place in the Intelligence Index with a cybersecurity-focused model
Gemini 4 Carbon: Google DeepMind reclaims the top spot on the Intelligence Index with a cybersecurity-focused AI model, two weeks after Gemini 4 Argon.
OpenAI GPT-6.1 Terra: the mid-range model that drives screens and apps, at half the price
OpenAI launches GPT-6.1 Terra, the mid-range agentic model that drives screens and apps at half the price. A deep dive for developers.
**GVS5H: five Qwen3.8-27B orchestrated via a shared folder beat Claude Fable 5 at 92.4% on LiveCodeBench Hard**
GVS5H: five Qwen3.8-27B instances orchestrated via a shared folder beat Claude Fable 5 with 92.4% on LiveCodeBench Hard. Discover this protocol.
Kimi K3.1 Pro: Moonshot overtakes GPT-6 Astra on Terminal-Bench 4.0 and opens up its 3 trillion parameters on November 4
Moonshot AI's Kimi K3.1 Pro surpasses GPT-6 Astra on Terminal-Bench 4.0: a 3-trillion-parameter open-weight model, available November 4.
Carbon-X: Hugging Face's open 14B model that "discusses" with DNA, 1,800 times cheaper than GPT-6.1 Sol
Carbon-X: Hugging Face's open 14B model that converses with DNA in natural language, 1,800 times cheaper than GPT-6.1 Sol.
Mercury 3: Inception's diffusion LLM claims to beat GPT-6 Luna and Claude Haiku 5.5 at over 1,000 tokens/s
Mercury 3 by Inception Labs: the diffusion LLM promises over 1,000 tokens/s and challenges GPT-6 Luna and Claude Haiku 5.5. Discover this revolutionary model.
NVIDIA Nemotron 3.5 Super VL: the open 120B VLM (12B active) targeting multimodal agents — and the Bolt 74B
NVIDIA Nemotron 3.5 Super VL: the open 120B-parameter VLM (12B active) built for multimodal agents, with compact models Bolt 74B and E4B.
Liquid AI enters the decision-making model market with d1-ultra
Liquid AI launches d1-ultra, a decision model that responds with probabilities instead of generating text. Discover this emerging new AI category.
Grok Voice 3.0: xAI achieves 98.5% voice reasoning and switches to grok-voice-latest
Grok Voice 3.0: xAI launches its real-time speech-to-speech model with 98.5% voice reasoning on Big Bench Audio, still priced at $0.08 per minute.
Meta launches Muse Wonder and Muse Wonder Full: its first frontier models, with active memory
Meta launches Muse Wonder and Muse Wonder Full, its first frontier models with active memory. Discover the capabilities of these new AI models.
GLM 5.3 Fast: Z.AI bets on speed and price to chip away at the LLM API mid-market
GLM 5.3 Fast from Z.AI bets on speed and price to conquer the LLM API mid-market. Analysis of a launch shaking up the fast segment.
**Best LLMs (October 2026)**
Discover the ranking of the best LLMs of October 2026: agentic capabilities, AI model comparison, and market shift analysis.
Best Local LLMs (October 2026)
Ranking of the best local LLMs in October 2026: open source models to run on your machine, performance and privacy.
Best Llm Code (October 2026)
Discover the best LLM for coding in October 2026: GPT-5.2 Codex, Claude Opus 4.5 or DeepSeek V3.2? Full comparison of performance and pricing.
Anthropic rounds out the Claude 5.5 family with Haiku 5.5: three models in one month and a 1 million token context window
Anthropic launches Claude Haiku 5.5, completing the Claude 5.5 family: three models in one month, with a 1 million token context window.
Mistral releases Large 4: 1 trillion open-weight parameters, Europe's bet on the frontier
Mistral Large 4: 1 trillion open-weight parameters, multimodal MoE architecture. Mistral AI's European bet on the AI frontier.
Aleph Alpha open-sources Kolibri-1: a bilingual DE/EN 78B MoE (3.46B active) under Apache 2.0 for sovereign on-premise AI
Aleph Alpha open-sources Kolibri-1: a bilingual DE/EN 78B MoE under Apache 2.0 for sovereign on-premise AI. Discover this German model.
GPT-6.1 Sol at 1/5 the price of Astra while Gemini 4 disappoints in real-world use: the market rationalizes
OpenAI launches GPT-6.1 Sol, nearly on par with Astra at 1/5 the price, while Gemini 4 disappoints in real-world use: the AI market is rationalizing.
Xiaomi MiMo-V2.6: Live-streamed RL training that propels an open-source model to #1 on Artificial Analysis
Discover Xiaomi MiMo-V2.6: live RL training, open-source models up to 1T parameters, #1 on Artificial Analysis. Weights available on Hugging Face.
Anthropic accuses GLM-5.3 of "Mythos-class" hacking capabilities — and reignites the war against open weights
Anthropic accuses GLM-5.3 of Mythos-class hacking capabilities and reignites the war against open weights models. Analysis of the explosive report.
MiniMax M3.1 Flash Preview: China continues its war of small, fast models
Discover MiniMax M3.1 Flash Preview, the new fast small Chinese model quietly launched on September 27, 2026, without a model card or press release.
Claude Sonnet 5.5: Anthropic targets the mid-market with a model that's 30% faster and 30% cheaper
Claude Sonnet 5.5: Anthropic targets the mid-market with a model 30% faster and 30% cheaper. An analysis of the new model in the Claude 5.5 family.
Naive-N0.5-Flash: the open-weight 309B MoE with no full-attention layers at all aims for 2,000 tokens/s per user
Naive-N0.5-Flash: NaiveAI's open-weight 309B-parameter MoE without full-attention aims for 2,000 tokens/s per user on Hugging Face.
Gemini 3.8 Live with Live Avatar: Google adds real-time visual presence to its conversational assistant
Google announces Gemini 3.8 Live with Live Avatar: a conversational assistant with real-time visual presence that listens, sees and speaks.
The price war: GPT-6 Sol and Luna at half price, dropped 90 minutes after Claude Opus 5.5
AI price war: OpenAI launches GPT-6 Sol and Luna at half price, just 90 minutes after Claude Opus 5.5's release. Full analysis.
Gemini 3.8 Live: Google's real-time voice at $1.38 per hour — background reasoning and tool execution without interrupting the conversation
Gemini 3.8 Live: Google's real-time AI voice at $1.38/hr. Background reasoning and tool execution without interrupting the conversation.
DeepSeek V4.1-Flash: MIT, 552B, a quarter of the KV-cache, and the end of the Pro model behind the deepseek-flash endpoint as of September 14
DeepSeek V4.1-Flash: MIT weights on Hugging Face, 552B parameters, KV-cache cut to a quarter, and the Pro model ending as of September 14. Full analysis.
OpenAI GPT-6 Astra: 64.6% on Terminal-Bench-Science and ARC-AGI-3 nearly complete, but its written reasoning is becoming harder to monitor
OpenAI GPT-6 Astra scores 64.6% on Terminal-Bench-Science and nears ARC-AGI-3, but its written reasoning becomes harder to monitor.
DeepSeek swaps V4-Pro for V4.1-Flash behind the same endpoint: why you should re-test your pipelines before September 14
DeepSeek replaces V4-Pro with V4.1-Flash behind the same endpoint: find out why you should re-test your pipelines before September 14.
Best Local LLMs (September 2026)
Discover the best local LLMs of September 2026: complete ranking to run AI locally with privacy and no subscription.
# Best Free LLMs (September 2026) Could you also share the body text you'd like translated? I'd be happy to translate it while keeping URLs, /article/xxx slugs, and tool names intact.
Discover the best free LLMs in September 2026: Gemini, DeepSeek, OpenRouter, and more. An honest, up-to-date ranking to avoid pitfalls.
**Best LLM Code Tools (September 2026)** That's the English translation of the title. If you have the full article text you'd like translated as well, feel free to paste it and I'll translate it while preserving URLs, slugs, and markdown formatting.
Check out the ranking of the best LLMs for coding in September 2026: BenchLM scores, comparison and guide to choosing the right AI model.
iFlytek releases Spark X2.5: two open-source edge models with 1M context tokens — and a 293B arriving on September 7
iFlytek releases Spark X2.5: two open-source edge models with 1M context tokens. Also discover the 293B model planned for Sept 7.
GLM-5.3-Flash and Qwen3.8-Flash-Next: China releases two Flash models on the same day — the price war enters the speed race
GLM-5.3-Flash and Qwen3.8-Flash-Next: discover how China accelerates the AI model price war with these two simultaneous launches.
July 17: Gemini 3.5 Pro and Shanghai's WAIC collide — the day AI officially goes bipolar
On July 17, 2026, the Gemini 3.5 Pro launch and Shanghai WAIC illustrate two opposing visions. Discover this key day for AI.
GPT-Live : OpenAI launches full-duplex voice — AI agents can finally listen and speak at the same time
OpenAI launches GPT-Live with full-duplex voice. Discover how AI agents can finally listen and speak at the same time.
Meta Muse Spark 1.1 : Meta launches its first paid model and enters the agentic coding battle
Discover Meta Muse Spark 1.1, Meta's first paid model. The giant enters the agentic coding battle and changes strategy.