GPT-4
Core Facts
The GPT-4 series uses the cl100k_base vocabulary (approximately 100,000 tokens), while the GPT-4o series upgrades to o200k_base (approximately 200,000 tokens).
GPT-5.5是OpenAI于2025年中期发布的模型，相比GPT-4o在推理深度和上下文理解方面有显著提升
GPT-4o is developed by OpenAI and has an advantage in tool integration scenarios thanks to its Function Calling capability.
Timeline (last 90 days)
Open-source large models allow users to freely download, modify, and use them commercially (subject to the respective licenses), unlike closed-source models such as GPT-4.
Gates' article "The Age of AI Has Begun" was published at almost the same time as OpenAI released GPT-4.
The paper reports that RAPTOR, paired with GPT-4 on the QuALITY benchmark (long-document question answering), improved absolute accuracy by about 20 points, reaching state of the art at the time.
The core capabilities of email AI assistants are typically built on large language models (LLMs), such as GPT-4.
Alpha Codium research shows that a closed-loop approach of "plan—write—run tests—see errors—fix—repeat" raises GPT-4's pass rate on the CodeContests benchmark from 19% to 44%.
The training clusters of almost all frontier models such as GPT-4, Llama 3 and Claude are centered on the H100.
The first wave of Agent frameworks mainly solved capability problems (chain of thought, tool use, multi-step planning), because capability performs excellently in demos, but the hard part was never capability—it was reliability, observability, and controllability.
Even GPT-4-level models cannot fundamentally eliminate hallucinations.
Open-weight models such as Llama, Mistral, and Qwen may lag behind top closed-source models such as GPT-4 or Claude by a few percentage points on benchmarks like MMLU and GPQA, but they already have practical value for common academic research tasks such as literature summarization, coding assistance, hypothesis generation, and data cleaning.
Local small models are weaker in capability than cloud large models such as GPT-4, especially at small parameter sizes, and are prone to omissions or misreadings in long-document understanding and complex legal reasoning.
40 more timeline events
All Facts (20)
The GPT-4 series uses the cl100k_base vocabulary (approximately 100,000 tokens), while the GPT-4o series upgrades to o200k_base (approximately 200,000 tokens).
90%VerifiedGPT-5.5是OpenAI于2025年中期发布的模型，相比GPT-4o在推理深度和上下文理解方面有显著提升
90%VerifiedGPT-4o is developed by OpenAI and has an advantage in tool integration scenarios thanks to its Function Calling capability.
90%VerifiedSince 2023, with the leap in reasoning capabilities of large language models such as GPT-4 and Claude, Agents have rapidly moved from academic concepts to engineering practice.
90%VerifiedGPT-4o, released in 2024, directly integrated image generation capabilities into a multimodal large language model.
90%VerifiedThe LLM-as-Judge method uses large language models such as GPT-4 as judges to automatically score the output quality of Agents.
90%VerifiedThe training cost of GPT-4 is estimated to exceed $100 million.
90%VerifiedThe pre-training process of large language models requires thousands of high-end GPUs running in parallel for months, with costs typically ranging from tens of millions to hundreds of millions of dollars.
90%VerifiedThe MoE architecture divides a model's internals into multiple expert sub-networks, with a gating network dynamically selecting a small number of experts to participate in computation during each inference, so that the actual number of activated parameters is far smaller than the total parameter count.
90%VerifiedRepresentative large language models such as GPT-4, Claude, and Gemini typically have parameter counts in the hundreds of billions or more.
90%VerifiedThe parameter count of GPT-4 is estimated to exceed 1 trillion
90%Verified早期GPT-3的上下文窗口仅有4K Token，而现代模型已扩展至128K甚至更长
90%VerifiedTraining a model at the level of GPT-4 requires tens of thousands of A100/H100 GPUs working in concert for several months.
90%VerifiedThe core working principle of large language models such as GPT-4, Claude, and Gemini is next-token prediction based on the Transformer architecture.
90%VerifiedGPT-4 was launched in 2023 with an estimated parameter scale exceeding one trillion.
90%VerifiedMultiple academic studies in 2024 showed that models such as GPT-4 are able to identify certain types of software vulnerabilities in controlled environments, with efficiency approaching that of a junior security researcher.
85%VerifiedGPT-4 and Claude are large language models that can understand context, generate structured outputs, and execute multi-step reasoning
85%VerifiedMixture of Experts (MoE) models dynamically select a small number of experts to activate for each input token through a gating mechanism, thereby reducing the computational cost per inference while maintaining model capability.
85%VerifiedIn 2023, thousands of AI researchers and tech leaders co-signed an open letter calling for a six-month pause on training AI systems more powerful than GPT-4.
85%VerifiedAzure OpenAI Service provides enterprise customers with the ability to invoke models such as GPT-4, DALL-E, and Whisper in a compliant environment.
85%