ChatGPT vs Claude: plans, prices and benchmarks in October 2026
Both charge $20 a month for the main paid plan. Artificial Analysis scores Claude Opus 5.5 at 58 and OpenAI's best listed model at 53.
770 stories
Both charge $20 a month for the main paid plan. Artificial Analysis scores Claude Opus 5.5 at 58 and OpenAI's best listed model at 53.
A new arXiv benchmark scores shopper simulators on 64,729 real sessions and checks one against a production A/B test: the best still understates the real effect by 28.7%.
The Free plan covers basic search and 3 Pro Searches a day. Pro is $20 a month and Max is $200, as of November 18, 2026.
OpenAI's docs point to the usage dashboard and two CLI commands. Here is what each shows and how to make your limits last.
Where to see your remaining Cursor usage on The X Subscription and on legacy Pro plans, as of October 27, 2026.
Claude's Free plan covers everyday chat, search and file creation with five-hour and weekly usage limits. Claude Code, Opus and Research need Pro.

CNBC reports the order book closes November 17 and Nasdaq trading as ANT is expected November 19; Anthropic has not announced the date itself.
155 long-prompt responses contained other customers' data between August 27 and September 9. The cause was a configuration mismatch and a bug in SGLang.
iOS 27.2 is out as of October 26. Apple's notes list Siri AI in five more languages, Shared Albums tools and Simple Profiles in the Apple TV app.
Settings > Usage in the app, /usage and /context in Claude Code, and the Console Usage page for the API, with what each number means as of October 26, 2026.
Both have a 1.05M-token context window. Sol leads on Artificial Analysis's index, 52 to 48, while Terra is cheaper per token and faster.
Oxford, NUS, NTU and UC Merced researchers sort video AI into four operations and argue that smooth video is not proof a model understands the world.
The SEC filings describe a mandatory convertible preferred and $22.5 billion of forward-sold common stock, with a $60 billion at-the-market program planned for 2027.
English, French, Japanese, Korean, Portuguese and Spanish work now. Four more come by year end, and the EU is still left out on iPhone and iPad.
The mid-size model has a 1.05M-token context window, is live in ChatGPT Work, Codex and the API, and costs half as much as GPT-5.6 Terra, OpenAI says.
The S-1 shows revenue of $16.3 billion in six months, an $80.7 billion net loss driven by non-cash charges, and founder voting control.

Nvidia says the three-model suite labels robot training data about 125 times cheaper than GPT-6 Astra, though it trails frontier models on some benchmarks.
Context is one chat's working memory; the usage limit is your plan's allowance over time. Here is how to check each.
Warp lists plans from $0 to $200 a month, but agent work is billed by usage, in dollars for some accounts and credits for others.
Nanbeige4.3-33B has a 512K context window and, on the lab's own tests, tops Qwen3.8-27B and Gemma4-31B on most coding-agent benchmarks.
The Apache 2.0 model activates 2.5B parameters per token and, the company says, leads open models under 50B on agentic coding benchmarks.
The AI drug developer is now valued at about $3 billion, and BMS also gets access to its Foundry platform for neuroscience programs.
The government puts in £100 million, OpenAI and Anthropic pledge up to £200 million mostly in model credits, and Google DeepMind adds researchers.

The full-duplex model listens and talks at once, runs on a MacBook Pro M4 Max, and is limited for now to a waitlist and a group of safety researchers.