Home/Blog/ChatGPT vs Claude vs Gemini vs Grok (novembro de 2026): comparação honesta
Comparação10 min de leitura

ChatGPT vs Claude vs Gemini vs Grok (novembro de 2026): comparação honesta

GPT-6 Astra vs Opus 5.5/Fable 5.1 vs Gemini 4 Argon/Carbon vs Grok 4.7: lineups, preços de API, dois índices independentes com custo por tarefa de agente e veredictos por trabalho. Snapshot de novembro de 2026.

Por ZNIX AI Editorial · Documentação de modelos de IA e guias de uso
Publicado em

ChatGPT vs Claude vs Gemini vs Grok (November 2026): The Honest Four-Way Comparison

The four consumer AI brands now map to very different model lineups: ChatGPT runs GPT-6 Astra plus the cheaper GPT-6.1 Sol/Terra/Luna tiers; Claude runs Opus 5.5, Fable 5.1, Sonnet 5.5 and Haiku 5.5; Gemini’s current line is Gemini 4 Argon (public) and Carbon (restricted); Grok is on 4.7 with the Grok Build agent. This comparison uses the Artificial Analysis indexes and vendor pricing as reported on November 3, 2026, cross-checked against our own model reviews on November 20, 2026. It is a dated editorial snapshot, not a live benchmark.

The Lineups at a Glance

BrandFlagshipValue tierAPI input/output (per 1M)
ChatGPT (OpenAI)GPT-6 Astra ($10/$50)GPT-6.1 Sol $2/$10 · Terra $1/$5 · Luna $0.10/$0.50Astra $10/$50; Sol $2/$10
Claude (Anthropic)Opus 5.5 ($4/$20); Fable 5.1 for hard reasoning ($10/$50)Sonnet 5.5 $2/$10 · Haiku 5.5Opus $4/$20; Fable cached input $0.25
Gemini (Google)Gemini 4 Carbon (restricted access)Argon $2/$10 intro (to Nov 30) · 3.8 Flash $0.75/$3.75Argon $2/$10 → $4/$20 Dec 1
Grok (xAI)Grok 4.7—$2/$6 (<200K prompt); $4/$12 at 200K+

Watch the fine print: OpenAI rates rise above 272K input tokens, Grok doubles at 200K, and Argon’s introductory price ends November 30, 2026. Sticker prices also exclude thinking tokens, cache writes and repeated agent turns.

Benchmarks: Two Indexes, Two Different Questions

Intelligence Index v5.0 (broad capability; index points, not percentages): Gemini 4 Carbon 59 > Claude Opus 5.5 58 > GPT-6 Astra 57 > GPT-6.1 Sol 55 > Sonnet 5.5 54 > Gemini 4 Argon 50 > Fable 5.1 49 > Grok 4.7 35. The leader is a restricted-access model; among broadly available models, Opus 5.5 tops this snapshot.

Coding Agent Index v2.0 (complete agent setups, with cost and time per task): Claude Code + Opus 5.5 leads at 62 ($11.42/task, 1.0 h); Codex + GPT-6 Astra scores 55 ($7.04, 24.8 min); Codex + GPT-6.1 Sol scores 54 at just $1.50/task; Grok Build + Grok 4.7 scores 47 ($9.48, 35.8 min). The index version changed in October (v1.5 → v2.0), so numbers are not comparable with summer lists.

Keep independent indexes separate from launch claims: our Astra review notes the Requesty index scored Astra at 61 — level with GPT-5.6 Sol — despite headline launch benchmarks. Different index, different question; neither is “the” truth.

ChatGPT: The Widest Ladder

OpenAI’s advantage is tier coverage. GPT-6 Astra (Sept 3) is the reasoning and agent flagship: 1.1M-token input context, 128K output, computer use nearly 2× faster than its predecessor, and the first model to cross OpenAI’s “Critical” cybersecurity threshold. GPT-6.1 Sol (Sept 29) delivers near-Astra coding at a fifth of the price; Terra (Oct 26) targets visual reasoning and computer use; Luna handles high-volume light tasks. Astra is available through the API, ChatGPT Plus/Pro/Business/Enterprise, Codex and AWS. The trade-off flagged in the system card: decreased monitorability and safety-monitor false positives that can interrupt legitimate work.

Claude: The Coding Default

Anthropic’s current recommendation is Opus 5.5 (Sept 22) for most work, with Fable 5.1 reserved for demanding reasoning. Fable leads SWE-bench Pro at 81.2% and offers cached input at $0.25/1M — a 4× cache advantage over Astra for repetitive engineering workloads. Claude Code + Opus 5.5 tops the Coding Agent Index, but at $11.42 per benchmark task it is the premium option; Sonnet 5.5 ($2/$10) is the cost lever. If your workload is high-volume coding with repeated context, cache economics often matter more than headline scores.

Gemini: Multimodal Reach, Access Caveat

Gemini 4 Argon (GA Oct 13) is the pick for research across documents, audio, video and images, with a unified input/output context budget. Carbon (announced Oct 28) posts the strongest independent scores — 59 on the Intelligence Index, 59 with Antigravity CLI on the coding index — but its rollout is restricted to cyber defenders and trusted partners, so most readers cannot use it yet. Argon’s $2/$10 introductory pricing ends November 30; budget for $4/$20 from December 1.

Grok: Ecosystem Play, Mind the Long-Prompt Rate

xAI positions Grok 4.7 for coding, agentic tasks and knowledge work inside the Grok ecosystem. On the November snapshot it trails the other three on both indexes (35 intelligence; Grok Build 47 coding), and its completed-task cost ($9.48) sits above Codex setups. Pricing is competitive below 200K prompt tokens ($2/$6) but doubles to $4/$12 across the whole request at 200K+ — material for long-context agent runs. Evaluate it as a complete setup (Grok Build) rather than a bare model if you already live in that ecosystem.

Verdict by Job

JobStart withWhy
Daily codingClaude Opus 5.5 or GPT-6.1 SolTop agent scores; Sol at $1.50/task is the value pick
Hard reasoning / researchGPT-6 Astra or Claude Fable 5.1Keep the premium only if accepted results improve
Cache-heavy engineeringClaude Fable 5.1$0.25 cached input, SWE-bench Pro 81.2%
Multimodal researchGemini 4 ArgonDocuments/audio/video in one context; Carbon restricted
High-volume light tasksGPT-6 Luna or Gemini 3.8 Flash$0.10/$0.50 and $0.75/$3.75 respectively
Already in Grok/X ecosystemGrok 4.7 + Grok BuildCompare completed-task cost before switching

For Creators: The Layer Above Your Video Tools

None of these four renders video. They sit upstream — scripting, shot-listing, prompt refinement — while models like Seedance 2.0 and Kling 2.6 do the rendering. A practical stack: a value-tier chat model (Sol / Sonnet / Argon) for iteration, a premium model only for the hard brief, and credit-based video models for output. Copy-ready shot language lives in our video prompts library, and the full November leaderboard is in our monthly snapshot.

Sources: Artificial Analysis Intelligence Index v5.0 / Coding Agent Index v2.0 (Nov 3, 2026 snapshot) and OpenAI/Anthropic/Google/xAI pricing pages, as compiled by divkix.me/blog/ai-models-compared-2026 (secondary source, checked 2026-11-20); Astra/Fable/GPT-5.6 details from our own reviews linked above. Index points are not percentages; access limits and effort configurations kept as published. Verify live pricing before committing.

Perguntas Frequentes

ChatGPT vs Claude vs Gemini vs Grok — which should I use?
By job: daily coding starts with Claude Opus 5.5 or GPT-6.1 Sol ($1.50 per benchmark task); hard reasoning with GPT-6 Astra or Fable 5.1; multimodal research with Gemini 4 Argon (Carbon is restricted); Grok 4.7 mainly if you already live in the Grok/X ecosystem. On the November 2026 snapshot Grok trails the other three on both independent indexes.
Grok Build vs Claude Code — which coding agent wins?
On the Coding Agent Index v2.0 (Nov 3, 2026), Claude Code + Opus 5.5 scores 62 at $11.42 per task; Grok Build + Grok 4.7 scores 47 at $9.48. Claude Code leads on score; Codex + GPT-6.1 Sol is the value pick at 54 points and $1.50 per task. These are benchmark averages — measure on your own repositories.
Which is cheapest per token?
Among flagships, GPT-6.1 Sol and Claude Sonnet 5.5 are both $2/$10 per million tokens; Grok 4.7 is $2/$6 below 200K prompt tokens but doubles to $4/$12 at 200K+; Gemini 4 Argon is $2/$10 introductory until November 30, then $4/$20. For light high-volume tasks, GPT-6 Luna ($0.10/$0.50) and Gemini 3.8 Flash ($0.75/$3.75) are cheaper still.
Can any of these four generate video?
No — they are language/reasoning models that sit upstream of video generation (scripting, shot lists, prompt refinement). Rendering is done by video models like Seedance 2.0, Kling 2.6 or Hailuo 2.3, available on ZNIX with credit-based pricing. Sora 2 is OpenAI’s video model but is separate from ChatGPT’s text lineup and has no API.

Continue lendo

Sobre o autor
ZNIX AI Editorial — Documentação de modelos de IA e guias de uso

A equipe editorial do ZNIX explica modelos de IA com fontes primárias, verificações de disponibilidade datadas e exemplos identificados. Resumos da documentação são diferenciados de testes práticos.

Pronto para criar vídeos com IA?

Ganhe 50 créditos grátis — sem cartão de crédito.

Comece Grátis →