AI News

Join 50k top AI Engineers every weekday.

We respect your privacy. Full signup link here.

We summarize top AI discords + AI reddits + AI X/Twitters, and send you a roundup each day!

"Highest-leverage 45 mins I spend everyday" - Soumith

" best AI newsletter atm " and " I'm not sure that enough people subscribe " - Andrej

"genuinely incredible" - Chris

"surprisingly decent" - Hamel

You can pay for a customizable version here . Thanks to Pieter Levels for the Lex Fridman feature!

Last 30 days in AI

Filter titles:

See all issues

Apr 25

Cognition's DeepWiki, a free encyclopedia of all GitHub repos

deepwiki perception-encoders qwen dia-1.6b deep-research grok-3-beta o4-mini gemini-2.5-pro claude-3.7 deepseek-r1-1776 openai-o4-mini vllm surya-ocr cognition meta alibaba hugging-face openai perplexity-ai vllm github ai-models vision-encoders chat-apps text-to-speech deep-research rlhf ocr datasets frameworks

AI news highlights the launch of DeepWiki, a free encyclopedia of GitHub repositories, and updates on various AI models and tools including Meta's perception encoders, Qwen chat app, Hugging Face integrations, OpenAI's deep research, and new OCR models. The news also covers model releases, inference capabilities, and new frameworks for AI development.
Apr 24

not much happened today

gpt-image-1 o3 o4-mini gpt-4.1 dam openai google anthropic epoch ai research image-generation model-benchmarks vision-language-models music-ai ai-experiences ai-research supercomputers

AI news for April 23-24, 2025, covering new model releases, benchmarks, and research developments from companies like openai, google deepmind, anthropic, and epoch ai research.
Apr 23

gpt-image-1 - ChatGPT's imagegen model, confusingly NOT 4o, now available in API

gpt-image-1 gpt-4 gpt-4-mini gpt-4.1 gemini-2.5-pro qwen-2.5-vl-72b eagle-2.5-8b dam openai nvidia image-generation multimodal model-performance benchmarking supercomputing content-moderation long-video-understanding ai-scaling virology-capabilities

Autoregressive Imagegen introduces a new official API for GPT 4o, enabling advanced image generation, editing, transparency, and moderation features. The news covers recent AI model performance, benchmarks, supercomputer scaling, and Nvidia's new multimodal model for image and video captioning.
Apr 22

not much happened today

nemotron-h eagle-2.5 gemini-2.5 gemma-3 qwen2.5-32b scaling01 uni3c seedream-3.0 kimina-prover bitnet-b1.58 the-ai-timeline nvidia alibaba bytedance adobe model-releases multimodal transformers long-context-learning reinforcement-learning video-generation diffusion-models large-language-models efficient-llms ai-research

A quiet day in AI news with updates on model releases including Nemotron-H, Nvidia Eagle 2.5, Gemini 2.5, Gemma 3, SRPO with Qwen2.5-32B, Alibaba's Uni3C, ByteDance's Seedream 3.0, Adobe DRAGON, Kimina-Prover, and BitNet b1.58, along with research on antidistillation sampling.
Apr 21

not much happened today; New email provider for AINews

gemini-2.5-pro gpt-4.1 gpt-4.1-mini gpt-4.1-nano gemini-2.5-flash seaweed-7b claude embed-4 grok openai google byte-dance anthropic cohere xai model-releases benchmarks reasoning-ai multimodal text-to-video video-ai audio-driven-synthesis values-in-ai embedding-models search-and-retrieval ai-automation agentic-workflows

AI news from April 18-21, 2025, covering model updates, new releases, and AI applications including OpenAI, Google Gemini, ByteDance Seaweed, Anthropic Claude, Cohere Embed 4, Elon Musk's xAI Grok, and AI workflows.
Apr 19

Grok 3 & 3-mini now API Available

grok-3-mini gemini-2-5-pro gemini-2-5-flash llama-3 maverick o3 o4-mini gemma-3 x.ai openai google deepmind llamaindex gdb mckbrando epochai-research goodfireai tamaybes reach_vb ai-agent-development multi-agent-systems ai-code-editing model-performance benchmarking ai-startups neural-programming virtual-work-environments quantization-aware-training

AI news from 4/17/2025-4/18/2025 covers new model releases like Grok 3 mini, Gemini 2.5 Pro and Flash, and updates on Llama 3, Maverick, and Gemma models. Highlights include AI agent frameworks, code editing tools, startup funding, and infrastructure advancements.
Apr 18

Gemini 2.5 Flash completes the total domination of the Pareto Frontier

gemini-2.5-flash o3 o4-mini google openai model-releases multimodal tool-use benchmarking ai-research ai-models

Gemini releases Gemini 2.5 Flash with new 'thinking budget' feature, emphasizing model control and performance. OpenAI launches o3 and o4-mini models with strong multimodal and tool use capabilities. The AI community discusses model benchmarks, tool integration, and Google's renewed focus on Gemini.
Apr 17

OpenAI o3, o4-mini, and Codex CLI

o3 o4-mini gpt-4.1 gemini-2.5-pro seedream-3.0 openai reinforcement-learning multimodal tool-use model-optimization open-source benchmarking ai-models

OpenAI launched o3 and o4-mini models with improvements in scaling RL and efficiency, featuring better vision and tool use, and introduced Codex CLI, an open-source coding agent. The release received positive feedback, with comparisons to other models like Gemini 2.5 Pro and benchmarks showing enhanced performance.
Apr 16

QwQ-32B claims to match DeepSeek R1-671B

qwen2.5-plus qwen-2.5-plus qwenq-32b gpt-4.5 alibaba openai reinforcement-learning multimodal model-optimization benchmarking ai-models language-models math coding alignment human-preference

The article discusses the release of Alibaba's Qwen2.5-Plus + Thinking (QwQ) model, a 32B reasoning model with staged reinforcement learning focusing on math, coding, and general capabilities. It also covers the rollout of GPT-4.5, user feedback, and benchmark comparisons.
Apr 16

SOTA Video Gen: Veo 2 and Kling 2 are GA for developers

gpt-4.1 veo-2 kling-2 openai google generative-ai video-generation api-updates model-performance benchmarking multimodal

AI news covers the release of GPT-4.1 family with improvements in coding, instruction following, and long context, along with new video generation models like Google's Veo 2 and Kling 2. The news also discusses API updates, model performance, and industry excitement.
Apr 15

GPT 4.1: The New OpenAI Workhorse

gpt-4-1 gpt-4-1-mini gpt-4-1-nano openai ai-models benchmarking prompting instruction-following coding api model-deprecation performance-comparison

GPT 4.1 from openai introduces new benchmarks, prompting guides, and improved performance in coding and instruction following. The release includes GPT-4.1 mini and nano models, with discussions on pricing, capabilities, and model deprecation. The news covers Twitter reactions, performance benchmarks, and industry implications.
Apr 11

not much happened today

grok-3 grok-3-mini gpt-4.5 claude-3.7-sonnet quasar-alpha optimus-alpha llama-4-scout llama-4-maverick internvl3 qwen2.5vl epochairesearch rasbt scaling01 sarahookr mervenoyann theturingpost cloneofsimo akhaliq gneubig lioronai openai svpino mathemagic1an language-models benchmarks reinforcement-learning multimodal vision-language-models agents tooling ai-infrastructure

AI news for April 10-11, 2025, covering language models, vision and multimodal models, AI agents, tooling, and infrastructure updates. Highlights include evaluations of Grok-3, RL improvements for small LLMs, new vision models like Kaleidoscope, InternVL3, and TransMamba, as well as AI applications such as FilmAgent AI and BrowseComp.
Apr 11

not much happened today

tpuv7 gb200 gpt-4.1 grok-3-mini ironwood-tpu google nvidia xai openai sama epoch-ai hardware-accelerators ai-models ai-hardware ai-industry ai-research ai-infrastructure ai-competitions ai-events

AI news from April 9-10, 2025, covering hardware updates, new models, and industry events. Highlights include Google's TPUv7, Ironwood TPU, GPT-4.1 upgrade leaks, X.ai's Grok 3 API, and various AI industry discussions.
Apr 10

Google's Agent2Agent Protocol (A2A)

gpt4o llama-4-scout llama-4-maverick deepcoder-14b llama-3-1-nemotron-ultra google deepmind anthropic meta uc berkeley nvidia agent-interoperability multimodal-models large-language-models coding-models ai-model-updates

Google Cloud Next announcements include full MCP support and new Agent to Agent protocol, enhancing agent interoperability with partners like Anthropic. The protocol features Agent Card, Task, Messages, Artifact, and supports enterprise auth, observability, streaming, and push notifications. Additionally, AI model updates include Moonshot AI's Kimi-VL-A3B, Meta's Llama 4 Scout and Maverick, DeepCoder 14B from UC Berkeley, and Nvidia's Llama 3.1 Nemotron Ultra 253B.
Apr 09

DeepCoder: A Fully Open-Source 14B Coder at O3-mini Level

gemini-2-5-pro kimi-vl-a3b deepcoder-14b llama-4-scout maverick gen-4-turbo imagen-3 veo-2 google meta moonshot-ai together-ai agentica huggingface model-releases multimodal text-to-image code-generation reinforcement-learning open-source vision-math-benchmarks

AI news covers open-source models, new model releases from Google, Meta, Moonshot AI, and collaborations like DeepCoder-14B by Together AI and Agentica. Highlights include Gemini 2.5 Pro, Kimi-VL-A3B multimodal LM, DeepCoder-14B coding model, Llama 4 Scout, Maverick, Google Imagen 3, and Veo 2.
Apr 08

Llama 4's Controversial Weekend Release

llama-4 meta large-language-models multimodal model-releases training-data benchmarking open-models

Meta released Llama 4, a new multimodal large language model, with mixed reception regarding its transparency, performance, and implementation issues. The release included a 109B parameter Scout model and a 2 trillion parameter behemoth, aiming to restore Meta's position in open AI models. Discussions highlight concerns about training data, benchmark performance, and release transparency.
Apr 05

not much happened today

o3 o4-mini gpt-5 gemma-3 qwen-2.5-vl gemini-2.5-pro gemma-2-27b llama-3.1 openai google anthropic meta deepseek model-releases model-capabilities benchmarks long-form-reasoning reward-modeling attention-sinks scaling large-language-models mixture-of-experts

AI news covers model release updates, new model capabilities, benchmarks, and industry developments including OpenAI, google, anthropic, meta, and deepseek. Highlights include GPT-5 delays, Gemini 2.5 Pro preview, and research on LLM attention behavior.
Apr 04

not much happened today

gemini-2-5-pro deepseek-v3 qwen-2-5 claude-3-7-sonnet google meta anthropic openai runway large-language-models model-performance ai-evaluation chains-of-thought ai-safety ai-tools ai-frameworks multi-agent-systems media-creation

AI news for April 2-3, 2025, covering model performance, new tools, and research insights. Highlights include Gemini 2.5 Pro, DeepSeek V3, Qwen 2.5, and concerns about chains-of-thought in LLMs, along with new frameworks like PaperBench and CodeAct.
Apr 02

not much happened today

gpt-2 gemma-3 gemmacoder3-12b qwen-2.5-omni gemini openai deepseek alibaba amazon bmw open-source-models model-performance benchmarks ai-product-releases humanoid-robots ai-research

AI news for late March 2025 covers open-source model releases, benchmark performances, new AI tools, and research updates, highlighting OpenAI's open-weight models, Gemini's math progress, and humanoid robot deployments.
Apr 01

>$41B raised today (OpenAI @ 300b, Cursor @ 9.5b, Etched @ 1.5b)

gpt-4 sonnet-3.5 o1 gemini-2.5-pro gemini-2.5 gemini-2.0 deepseek-v3-0324 amazon openai deepseek scaling sophont gemini skypilot agentevals karpathy language-models open-models multimodal healthcare-ai ai-application ai-frameworks ai-tools cost-efficiency ai-research ai-startups

AI News for 3/28/2025-3/31/2025 covers major funding rounds, OpenAI's upcoming open language model, advancements in Gemini models, new multimodal healthcare foundation models, and AI tools like SkyPilot and AgentEvals.
Mar 28

not much happened today

gpt-4o deepseek-v3 claude-3-sonnet o3-mini gemini-2.5-pro sama kevinweil artificialanlys joannejang nrehiew_ giffmana philschmid scaling01 cursor_ai windsurf_ai saranormous togethercompute hypertecgroup coreweave lateinteraction model-performance coding-assessment instruction-following policy-compliance image-generation multimodal video-understanding audio-processing ai-infrastructure gpu-clusters ipo ai-engineering

AI news for March 27-28, 2025, covering model updates, performance assessments, infrastructure developments, and AI engineering surveys.
Mar 28

not much happened today

gpt-4o gpt-4o-latest gemini-2.5-pro gemini-3 openai swyx abacaj nrehiew_ sama joannejang giffmana lmarena_ai artificialanlys sambanova multimodal image-generation model-performance api ai-models ai-policy benchmarking coding-ai

AI news for 3/26/2025-3/27/2025 covers updates on GPT-4o, multimodal models, image generation, model performance, DeepSeek APIs, and Gemini models, highlighting recent advancements, policy considerations, and new releases.
Mar 27

OpenAI adopts MCP

gemini-2.5-pro qwen-2.5-omni-7b deepseek-v3-0324 openai google alibaba mcp-support large-language-models multimodal-ai model-performance benchmarking synthetic-data model-speed model-quantization

OpenAI announced MCP support, highlighting advancements in AI model interoperability. The news covers recent model performance updates, including Gemini 2.5 Pro's top rankings, Qwen 2.5 Omni's multimodal capabilities, and DeepSeekV3's benchmark improvements. Discussions include scaling laws of synthetic data, model speed, and benchmark reliability.
Mar 26

Gemini 2.5 Pro + 4o Native Image Gen

gemini-2.5-pro gpt-4o google openai multimodal autoregressive image-generation language-models ai-research

AI news highlights the release of Gemini 2.5 Pro, the top model in the world with significant improvements and involvement from Noam Shazeer, and OpenAI's release of GPT 4o native image generation, a new autoregressive model with advanced multimodal capabilities.
Mar 25

Halfmoon is Reve Image: a new SOTA Image Model from ex-Adobe/Stability trio

reve stability adobe text-to-image visual-generation natural-language-understanding interactive-systems prompt-engineering

Reve, a new image generation model from former Stability and Adobe alumni, aims to enhance visual generative models with logic and understanding of user intent, moving beyond simple slice-of-the-world generation.
Mar 22

lots of little things happened this week

claude-code llama-3.3-nemotron-49b sakata-ai gpt-4 gpt-3 claude-2 anthropic gemini nvidia meta roboflow kyutai-labs topaz-labs percy-liang instruction-following multi-step-problem-solving reasoning-benchmarks ai-models ai-research reinforcement-learning multiturn-agents ai-benchmarks ai-competitions

AI news highlights from March 20-21, 2025, including launches, research, benchmarks, and new models from companies like anthropic, gemini, nvidia, and meta, as well as updates on reasoning benchmarks and tools.
Mar 20

Promptable Prosody, SOTA ASR, and Semantic VAD: OpenAI revamps Voice AI

gpt-4o-transcribe gpt-4o-mini-tts o1-pro openai voice-models speech-to-text text-to-speech realtime-voice-detection audio-models model-performance open-source

OpenAI has launched new voice models including speech-to-text, text-to-speech, and real-time voice detection updates, along with a demo site and new API features. The community is excited about advancements in audio processing, model performance, and open-source initiatives.
Mar 20

Every 7 Months: The Moore's Law for Agent Autonomy

cosmos-transfer1 gr00t-n1-2b llama-4 phi-4-multimodal claude-3.7-sonnet nvidia meta microsoft agent-autonomy multimodal robot-reasoning text-to-speech ai-scaling-laws

AI news from March 18-19, 2025, highlights include research on agent autonomy scaling laws, Nvidia model releases Cosmos-Transfer1, GR00T-N1-2B, and Orpheus 3B, Meta's Llama-4 delay, and Microsoft's Phi-4-multimodal launch.
Mar 18

not much happened today

gemini-2-0-flash imagen-3 mistral-small-3.1 olmo-32b shieldgemma-2 google mistral-ai allen-ai langchainai abacaj qodoai stripe-dev multimodal language-models open-source-llm ai-frameworks ai-safety ai-tools ai-applications

Nvidia GTC day features announcements including Google's Gemini 2.0 Flash with multimodal capabilities, Mistral AI's Small 3.1 with expanded context, Allen AI's OLMo-32B outperforming some models, ShieldGemma 2 for image safety, and updates on frameworks like LangChainAI and fasttransform. The event highlights advancements in multimodal models, open-source LLMs, and AI tools.
Mar 18

Cohere's Command A claims #3 open model spot (after DeepSeek and Gemma)

cohere-command-a-111b mistral-ai-small-3.1 smol-docling cohere mistral-ai smol-ai hugging-face large-language-models multimodal multilingual benchmarking open-source context-windows ocr model-ranking

AI news from March 14-17, 2025, highlights open weights models, including Cohere's Command A, Mistral AI Small 3.1, and SmolDocling OCR. Notable discussions include model rankings, benchmarks like mcbench and HCAST, and the focus on long context windows, multilingual capabilities, and open-source deployment.

See all issues

Let's Connect

If you want to get in touch with me about something or just to say hi, reach out on social media or send me an email.