ChatGPT vs Claude vs Gemini in October 2026: models, prices and benchmarks compared

Which model actually powers each assistant right now

OpenAI's consumer chatbot is built on the GPT-6 family. GPT-6 Astra, released to limited organizations on September 4, 2026 and to Plus, Pro, Business and Enterprise ChatGPT tiers over the following days, is OpenAI's flagship reasoning model for coding, computer use and multi-step analysis, according to a report by 9to5Mac. Day-to-day chat in the app runs on GPT-6 Sol and GPT-6 Luna, which OpenAI rolled out in October 2026 to replace the earlier GPT-5.6 Sol and Luna versions across free and paid plans, per OpenAI's own Deployment Safety Hub documentation.

Anthropic's current flagship is Claude Opus 5, launched July 24, 2026. Anthropic describes it as close to the capability of the costlier Fable 5 at roughly half the price, while it remains behind the specialised Mythos 5 on cybersecurity and biological-research tasks, the company said in its launch announcement.

Google's broadly available flagship for Gemini app users is Gemini 3.1 Pro, an update within the Gemini 3 family that Google first introduced on November 18, 2025, according to pricing and comparison pages from nocode.mba, eesel.ai and tech-insider.org. A newer model, Gemini 4 Argon, was announced September 30, 2026, but Google said it is rolling out gradually, starting with a 'Fairwind Program' of cyber-defence testers before reaching paid API customers and Google AI Ultra subscribers.

Consumer pricing side by side

Only Plus and Pro/Business ChatGPT subscribers get GPT-6 Astra; the Free and Go tiers do not include it, according to a pricing breakdown by eesel.ai. Claude Pro and above include Opus 5, while the free Claude tier is limited to Sonnet and Haiku-class models, per Anthropic's own pricing page.

API cost and context window

Google's own benchmark figures for Gemini 4 Argon include 77.9% on the DeepSWE v1.1 coding test and a 68% score on CWE-bench v1 for security-vulnerability fixes. Anthropic says Opus 5 reaches 'state-of-the-art' results on its Frontier-Bench and GDPval-AA evaluations and roughly three times the score of the next-best model on the ARC-AGI 3 reasoning test, while OpenAI reports GPT-6 Astra scoring 98% on FrontierMath Tier 4 and 99.9% on ARC-AGI-3. Because each company runs and reports its own benchmarks, these figures are not directly comparable across vendors.

Older-but-still-current models used for most cross-vendor comparisons

Because Gemini 4 Argon and GPT-6 Astra are either newly released or still in limited rollout, most third-party comparisons published in 2026 still benchmark the prior generation — GPT-5.5, Claude Opus 4.8/4.5 and Gemini 3.1 Pro — which remain what most subscribers actually use day to day. One such comparison, published by tech-insider.org, put GPT-5.5 and Claude Opus 4.8 within a point of each other on SWE-bench Verified coding tests (88.7% vs 88.6%), with Gemini 3.1 Pro behind at 80.6%, while Gemini led on API price at $2/$12 per million tokens versus $5/$25–30 for the other two.

How the three differ beyond the numbers

• OpenAI iterates fastest and names models separately for reasoning (Astra-class) versus everyday chat (Sol/Luna-class), so ChatGPT users may be switched between models depending on the question.
• Anthropic markets Claude around agentic coding and long-running tasks, and lets users set an 'effort' level to trade cost for capability on Opus 5.
• Google bundles its AI subscriptions with cloud storage, YouTube Premium and Workspace features, and currently offers the cheapest entry-level paid tier at $4.99/mo (AI Plus).
• Google is the only one of the three currently restricting its newest flagship (Gemini 4 Argon) to a limited tester program rather than shipping it broadly on day one.

Which assistant fits which need

Readers choosing between the three are, in effect, choosing between OpenAI's broad reach and fast iteration, Anthropic's focus on coding and agentic reliability, and Google's integration with Search, Workspace and Android plus the lowest-cost entry tier. None of the companies' own benchmark claims are independently audited, so the comparisons above should be read as company-reported figures rather than neutral rankings.