Frontier Log

Ten LLM Capabilities Every Frontier Lab Now Offers

by Zhu Liang

Of the ten capabilities OpenAI, Anthropic and Google all now offer, OpenAI shipped four first. Anthropic and Google shipped three each.

Ten stacked layers, one per capability, each starting on the month the first lab offered it and coloured by that lab: Google's web search (Mar 2023), Anthropic's 100K-token context (May 2023), OpenAI's function calling (Jun 2023), Google's image input (Jul 2023), Google's 1M-token context (May 2024), OpenAI's structured outputs (Aug 2024), memory and reasoning (Sep 2024), Anthropic's computer use (Oct 2024) and MCP (Nov 2024).
Each layer starts when the first of OpenAI, Anthropic and Google opened the capability to use or sign-up, previews and betas included, from web search in Bard in Mar 2023 to MCP in Nov 2024. Download
CapabilityFirst to shipOpenAIAnthropicGoogle
Web searchGoogle, BardMar 23, 2023Mar 20, 2025Mar 21, 2023
100K-token contextAnthropic, ClaudeNov 6, 2023May 11, 2023Feb 15, 2024
Function callingOpenAI, GPT-4 and GPT-3.5 Turbo APIJun 13, 2023Nov 21, 2023Dec 13, 2023
Image inputGoogle, Bard, with Google LensSep 25, 2023Mar 4, 2024Jul 13, 2023
1M-token contextGoogle, Gemini 1.5 Pro and 1.5 FlashApr 14, 2025Aug 12, 2025May 14, 2024
Structured outputsOpenAI, its APIAug 6, 2024Nov 14, 2025Nov 5, 2025
MemoryOpenAI, ChatGPTSep 5, 2024Sep 11, 2025Aug 13, 2025
ReasoningOpenAI, o1-preview and o1-miniSep 12, 2024Feb 24, 2025Dec 20, 2024
Computer useAnthropic, Claude 3.5 Sonnet APIJan 23, 2025Oct 22, 2024Oct 7, 2025
MCPAnthropic, Claude DesktopMay 21, 2025Nov 25, 2024Jun 25, 2025

OpenAI, Anthropic and Google all now offer ten capabilities that each began at one of them, according to the labs' own posts. OpenAI was first on four, Anthropic on three and Google on three. The earliest was web search: Google opened Bard, which it said draws on information from the web, to sign-ups on Mar 21, 2023. The latest was MCP, which Anthropic introduced with support in the Claude Desktop app on Nov 25, 2024.

OpenAI first on four

OpenAI added function calling to its API on Jun 13, 2023, and introduced Structured Outputs on Aug 6, 2024. It made memory in ChatGPT available to Free, Plus, Team and Enterprise users on Sep 5, 2024, after a test with a small portion of users, and released o1-preview and o1-mini in ChatGPT on Sep 12, 2024.

Anthropic and Google on three each

Anthropic expanded Claude to 100K tokens of context on May 11, 2023, offered computer use in the API in public beta on Oct 22, 2024, and introduced MCP. Google was first with web search in Bard, then let Bard take images in prompts on Jul 13, 2023, and put Gemini 1.5 Pro and 1.5 Flash in public preview with a 1M-token window on May 14, 2024.

Five months to two years for the rest

Reasoning reached all three labs fastest: Anthropic released Claude 3.7 Sonnet, with extended thinking, on Feb 24, 2025, about five months after o1-preview. Web search took longest. Anthropic added it to Claude on Mar 20, 2025, two years after Bard. The last of the ten to reach all three was structured outputs, which Anthropic added to the Claude API in public beta on Nov 14, 2025. The timeline below lists each lab's date and post for every capability.

How the dates are set

Each date is the day a lab's own post opened the capability to anyone to use or sign up for, previews and betas included. A feature for invited testers or one partner counts from when it opened wider. Google's structured outputs post, on Nov 5, 2025, expands support already in the Gemini API, and its posts don't date the start.

Timeline

  1. Google opens Bard, which draws on information from the web, to sign-upsGoogle
  2. OpenAI opens a web-browsing plugin for ChatGPT, in alpha from its waitlistOpenAI
  3. Anthropic expands Claude's context window to 100K tokensAnthropic
  4. OpenAI adds function calling to gpt-4-0613 and gpt-3.5-turbo-0613 in the APIOpenAI
  5. Google lets Bard take images in promptsGoogle
  6. OpenAI rolls out image input in ChatGPT to Plus and Enterprise usersOpenAI
  7. OpenAI previews GPT-4 Turbo with a 128K context windowOpenAI
  8. Anthropic adds tool use to the Claude API in beta, with Claude 2.1Anthropic
  9. Google brings Gemini Pro to the Gemini API with function callingGoogle
  10. Google opens sign-ups for Gemini 1.5 Pro, with a 128,000-token context windowGoogle
  11. Anthropic launches the Claude 3 models, with visionAnthropic
  12. Google puts Gemini 1.5 Pro and 1.5 Flash in public preview with a 1M-token windowGoogle
  13. OpenAI introduces Structured Outputs in the APIOpenAI
  14. OpenAI makes memory in ChatGPT available to Free, Plus, Team and Enterprise usersOpenAI
  15. OpenAI releases o1-preview and o1-mini in ChatGPTOpenAI
  16. Anthropic offers computer use in the API, in public betaAnthropic
  17. Anthropic introduces MCP, with local MCP servers in the Claude Desktop appAnthropic
  18. Google introduces an experimental thinking version of Gemini 2.0 for developersGoogle
  19. OpenAI launches Operator for Pro users in the USOpenAI
  20. Anthropic releases Claude 3.7 Sonnet, with extended thinkingAnthropic
  21. Anthropic adds web search to Claude, in preview for paid users in the USAnthropic
  22. OpenAI releases GPT-4.1 in the API, with up to 1M tokens of contextOpenAI
  23. OpenAI adds remote MCP servers to the Responses APIOpenAI
  24. Google launches Gemini CLI, with built-in MCP supportGoogle
  25. Anthropic gives Claude Sonnet 4 a 1M-token window, in public betaAnthropic
  26. Google lets the Gemini app reference past chatsGoogle
  27. Anthropic brings memory to Claude for Team and EnterpriseAnthropic
  28. Google releases the Gemini 2.5 Computer Use model in the Gemini APIGoogle
  29. Google expands Structured Outputs in the Gemini API, describing support already thereGoogle
  30. Anthropic adds structured outputs to the Claude API, in public betaAnthropic

Sources

Related

Events

Tags