| Zhipu GLM |
GLM-4 |
2024-01-16 |
128K |
Zhipu's API flagship with 128K context. All Tools can automatically choose browsing, Python, images, and custom functions. |
| Gemini |
Gemini 1.0 Pro |
2023-12-06 |
32768 |
Google's first public Gemini generation, natively multimodal. Pro targeted developers and the Bard upgrade. |
| Qwen |
Qwen-72B |
2023-11-30 |
32768 |
An early large open-source flagship. Chinese, English, and long-text ability were clearly stronger than 7B. |
| DeepSeek |
DeepSeek LLM |
2023-11-29 |
4096 |
7B/67B general dense models. DeepSeek's first base as it moved from code into general chat. |
| Anthropic |
Claude 2.1 |
2023-11-21 |
200K |
Doubled context to 200K tokens, reduced long-document hallucinations, and improved tool-use reliability. |
| OpenAI |
GPT-4 Turbo |
2023-11-06 |
128K |
Announced at the first DevDay with 128K context, JSON Mode, and lower pricing. The Assistants API and GPTs launched alongside it. |
| xAI |
Grok-1 |
2023-11-03 |
8192 |
xAI's first chat model opened to X Premium users. In March 2024 it open-sourced 314B MoE weights. |
| DeepSeek |
DeepSeek Coder |
2023-11-02 |
16384 |
DeepSeek's first public model, focused on code completion and repo understanding. It started the later Coder line. |
| Zhipu GLM |
ChatGLM3-6B |
2023-10-27 |
32768 |
Added function calling, a code interpreter, and agent tasks. Open-source small models started to gain tool use. |
| Kimi |
Kimi Chat |
2023-10-09 |
128K |
Moonshot's first long-context assistant, known for reading very long documents. In March 2024 it expanded further to about 2 million Chinese characters. |
| Qwen |
Qwen-7B |
2023-08-03 |
8192 |
The starting point of Qwen open source. A 7B dense model that led the later Qwen open-source family. |
| Anthropic |
Claude 2 |
2023-07-11 |
100K |
Brought context to 100K tokens. Writing and coding were clearly stronger than Claude 1, and it was an early flagship for long-document analysis. |
| Zhipu GLM |
ChatGLM2-6B |
2023-06-25 |
32768 |
Open-source context rose from 2K to 32K, with chat quality and inference efficiency both improved. |
| OpenAI |
GPT-4 |
2023-03-14 |
8192 |
OpenAI's first large-scale multimodal flagship. Accepts text and image input, with reasoning and instruction following clearly better than GPT-3.5. Offered 8K and 32K context at launch. |
| OpenAI |
GPT-4-32K |
2023-03-14 |
32768 |
The 32K-context version of GPT-4 for longer documents and multi-turn complex tasks. Later replaced by long-context models such as GPT-4 Turbo and GPT-4o. |
| Anthropic |
Claude 1 |
2023-03-14 |
9K |
Anthropic's first public chat model, emphasizing Constitutional AI safety alignment. A faster Claude Instant shipped at the same time. |
| Zhipu GLM |
ChatGLM-6B |
2023-03-14 |
2048 |
A bilingual Chinese-English chat model that can run on consumer GPUs. It moved GLM from papers into a usable product. |
| OpenAI |
GPT-3.5 Turbo |
2023-03-01 |
4096 |
The first Chat Completions API chat model and a sibling of early ChatGPT. Low cost and low latency, suited to everyday conversation and simple tasks. |
