{"id":3868,"date":"2026-07-28T05:13:04","date_gmt":"2026-07-28T05:13:04","guid":{"rendered":"https:\/\/aiopsschool.com\/blog\/?p=3868"},"modified":"2026-07-28T05:13:06","modified_gmt":"2026-07-28T05:13:06","slug":"the-master-guide-to-chatgpt-openai-codex","status":"publish","type":"post","link":"http:\/\/aiopsschool.com\/blog\/the-master-guide-to-chatgpt-openai-codex\/","title":{"rendered":"The Master Guide to ChatGPT & OpenAI Codex"},"content":{"rendered":"\n

Gold & Diamond Standard Edition \u2014 Verified against live sources, July 28, 2026<\/h3>\n\n\n\n
\n

How this guide was built:<\/strong> Compiled on July 28, 2026 from OpenAI’s official documentation (developers.openai.com\/codex \u2014 fetched live), OpenAI’s help center and pricing pages via current trackers verified July 22\u201328, and the Codex changelog. One honesty note: I’m Claude (built by Anthropic), so this is an outside-in synthesis of the public record on a competitor’s products \u2014 sourced carefully, but always confirm plan\/pricing details at openai.com before spending money. OpenAI ships changes weekly; treat this as an accurate snapshot.<\/p>\n<\/blockquote>\n\n\n\n

\n\n\n\n

PART 0 \u2014 THE 30-SECOND MENTAL MODEL<\/h1>\n\n\n\n

ChatGPT in 2026 is no longer a chatbot \u2014 it’s a platform<\/strong>: chat, agents, research, coding, browsing, voice, video, memory, and 60+ app connectors under one login. Codex is the coding-and-computer-use agent<\/strong> inside that platform, spanning CLI, IDE, desktop app, web\/cloud, GitHub, Slack, and mobile \u2014 all sharing one agent harness.<\/p>\n\n\n\n

The two skills that unlock everything: (1) route the right model\/tool to the right task<\/strong> (the Auto selector won’t always do it for you), and (2) delegate outcomes with verification<\/strong>, not step-by-step instructions.<\/p>\n\n\n\n

\n\n\n\n

PART 1 \u2014 THE MODEL FAMILY (Deep Dive, July 2026)<\/h1>\n\n\n\n

1.1 Current lineup at a glance<\/h2>\n\n\n\n
Model<\/th>Where it lives<\/th>API price (per 1M in\/out)<\/th>Context<\/th>Best for<\/th><\/tr><\/thead>
GPT-5.6 Sol<\/strong><\/td>Flagship. ChatGPT paid tiers (reasoning settings), Codex, API (gpt-5.6<\/code> alias \u2192 gpt-5.6-sol<\/code>)<\/td>$5 \/ $30 (cached input $0.50)<\/td>1.05M, 128K out<\/td>Hard reasoning, coding, cybersecurity, agentic tasks<\/td><\/tr>
GPT-5.6 Sol Pro<\/strong><\/td>ChatGPT Pro & Enterprise only<\/td>\u2014 (product feature)<\/td>\u2014<\/td>Higher-effort\/accuracy variant of Sol<\/td><\/tr>
GPT-5.6 Terra<\/strong><\/td>Work, Codex, API<\/td>$2.50 \/ $15 (cached $0.25)<\/td>1.05M<\/td>The balanced mid-tier<\/td><\/tr>
GPT-5.6 Luna<\/strong><\/td>Work, Codex, API<\/td>$1 \/ $6 (cached $0.10)<\/td>1.05M<\/td>High-volume, budget agents<\/td><\/tr>
GPT-5.5 \/ 5.5 Instant<\/strong><\/td>ChatGPT everyday default; Codex recommended model<\/td>~$5 \/ $30 class<\/td>1M<\/td>Fast daily chat; general Codex work<\/td><\/tr>
GPT-5.5 Pro<\/strong><\/td>Pro tiers, Codex heavy tasks<\/td>premium<\/td>\u2014<\/td>Deepest reasoning runs<\/td><\/tr>
GPT-5.3-Codex-Spark<\/strong><\/td>Codex, research preview (Pro)<\/td>\u2014<\/td>\u2014<\/td>Extra-fast lightweight coding turns<\/td><\/tr>
gpt-image-2 \/ Sora<\/strong><\/td>Images \/ video<\/td>usage-based<\/td>\u2014<\/td>Image & video generation<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

Timeline that matters: GPT-5.5 launched April 23, 2026 as OpenAI’s fully retrained agentic-first model and became Codex’s default. GPT-5.6 (Sol\/Terra\/Luna) launched July 9, 2026<\/strong> after a June 26 preview \u2014 all three share a 1.05M-token context and 128K max output. Legacy models (GPT-4o, GPT-4.1, o4-mini, original GPT-5) were retired from ChatGPT on February 13, 2026. The o3 reasoning line still exists separately for some workloads.<\/p>\n\n\n\n

1.2 Which model for what \u2014 the routing table<\/h2>\n\n\n\n
    \n
  • Everyday chat, quick answers, drafting<\/strong>\u00a0\u2192 GPT-5.5 Instant (it’s the default for a reason \u2014 fast and cheap for OpenAI, snappy for you).<\/li>\n\n\n\n
  • Hard reasoning, complex code, multi-step agent runs<\/strong>\u00a0\u2192 GPT-5.6 Sol (pick it explicitly via reasoning settings\/model picker; see the Auto-router warning below).<\/li>\n\n\n\n
  • Maximum accuracy, worth the wait<\/strong>\u00a0\u2192 Sol Pro (Pro\/Enterprise) or GPT-5.5 Pro.<\/li>\n\n\n\n
  • Production API at scale<\/strong>\u00a0\u2192 Terra first; escalate to Sol on evidence, drop to Luna for classification\/extraction\/routing volume.<\/li>\n\n\n\n
  • Rapid-fire small coding edits in Codex<\/strong>\u00a0\u2192 GPT-5.3-Codex-Spark (Pro preview).<\/li>\n\n\n\n
  • Set\u00a0effort level<\/strong>\u00a0per model where offered \u2014 low effort for simple tasks is a real cost\/speed lever; reasoning tokens bill as output.<\/li>\n<\/ul>\n\n\n\n

    1.3 \u26a0\ufe0f The Auto-router caveat (the most underrated fact in this guide)<\/h2>\n\n\n\n

    In ChatGPT’s default Auto<\/strong> mode, OpenAI silently routes each message to Sol, Terra, Luna, or an older GPT-5.5 variant based on undocumented logic. On a hard prompt you get the flagship; on a routine one you may not \u2014 and the UI doesn’t show which model answered unless you dig into settings (API users always get the model ID in metadata). Power-user rule: for anything that matters, pick the model explicitly instead of trusting Auto.<\/strong> This is also why two people on the same $20 plan can have wildly different quality experiences.<\/p>\n\n\n\n

    \n\n\n\n

    PART 2 \u2014 CHATGPT: PLANS, AND EVERY FEATURE THAT MATTERS<\/h1>\n\n\n\n

    2.1 Plans (July 2026)<\/h2>\n\n\n\n
    Plan<\/th>Price<\/th>The essentials<\/th><\/tr><\/thead>
    Free<\/strong><\/td>$0<\/td>Limited GPT-5.5, unlimited mini-class, standard voice, Study Mode. Ads shown (US).<\/td><\/tr>
    Go<\/strong><\/td>$8\/mo (global since Jan 2026)<\/td>~10x Free limits, longer memory, 3 Tasks. Still ad-supported; no<\/strong> Deep Research\/Agent Mode.<\/td><\/tr>
    Plus<\/strong><\/td>$20\/mo<\/td>The knowledge-worker sweet spot: GPT-5.6 Sol via reasoning settings, no ads, Deep Research (~10\/mo), Agent Mode (~40 msgs\/mo), Tasks (5 active), Advanced Voice, Sora video, Codex, Projects (~40 files), 60+ connectors, custom MCP.<\/td><\/tr>
    Pro $100<\/strong><\/td>$100\/mo (new April 9, 2026)<\/td>~5x Plus usage, Sol Pro, big Codex capacity. No unlimited Sora.<\/td><\/tr>
    Pro $200<\/strong><\/td>$200\/mo<\/td>Everything + unlimited Sora, Operator agent, highest limits. The Sora gap is the main reason to pay the extra $100.<\/td><\/tr>
    Business<\/strong><\/td>~$20\u201330\/seat<\/td>Shared workspaces, admin controls, ChatGPT Work.<\/td><\/tr>
    Enterprise<\/strong><\/td>Custom<\/td>Compliance, SSO, and Lockdown Mode<\/strong> (disables browsing\/Deep Research\/Agent Mode\/downloads for data containment).<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

    Rollouts vary by account and region \u2014 check your own model picker before upgrading. Codex Mobile is free on every<\/strong> plan (even Free\/Go) since May 14, 2026.<\/p>\n\n\n\n

    2.2 The feature arsenal \u2014 and when to reach for each<\/h2>\n\n\n\n

    Memory + Projects (the highest-leverage combo).<\/strong> Memory (rebuilt as “Dreaming V3,” June 2026) holds durable facts about you<\/em> \u2014 role, preferences, style. Projects<\/strong> are persistent workspaces holding context about a workstream<\/em> \u2014 files (~40 on Plus), links, standing instructions, and automatic within-project memory across sessions. The canonical setup: Memory knows your role; each Project knows the client\/codebase\/course. Review Memory periodically \u2014 it’s imperfect.<\/p>\n\n\n\n

    Search vs. Deep Research.<\/strong> Search = quick current facts. Deep Research<\/strong> = a multi-step agent that plans (you can edit the plan before it runs), issues sequential queries, reads dozens of pages over 5\u201330 minutes, and returns a cited report. Since Feb 2026 it connects to any MCP server<\/strong>, so it can research your internal wikis and datastores alongside the web. Caveat: it synthesizes, it doesn’t verify \u2014 check citations on anything consequential.<\/p>\n\n\n\n

    Agent Mode \/ Operator.<\/strong> Agent Mode (Plus: ~40 messages\/mo) hands ChatGPT a browser for multi-step web tasks \u2014 form filling, comparisons, bookings. Operator is the heavier-duty version on Pro $200.<\/p>\n\n\n\n

    Tasks (scheduled automation).<\/strong> One-off or recurring runs \u2014 daily briefings, inbox checks, monitors \u2014 without an open tab. Replaced the Pulse feature (sunset June 18, 2026). Active-task caps by tier (Go 3, Plus 5, higher tiers more).<\/p>\n\n\n\n

    Canvas.<\/strong> A live side-by-side document\/code editor for long-form iteration \u2014 edit sections, get inline suggestions, run code. Use it whenever output exceeds a few paragraphs.<\/p>\n\n\n\n

    Data Analysis.<\/strong> Upload CSV\/Excel; ChatGPT writes and runs Python for real calculations, charts, and cleaning. Also dedicated ChatGPT-in-Excel\/Sheets integrations.<\/p>\n\n\n\n

    Voice, Record Mode, images, Sora.<\/strong> Advanced Voice for natural conversation; Record Mode transcribes\/summarizes meetings; image input for screenshots; image generation for mockups\/diagrams; Sora for video (unlimited only on Pro $200).<\/p>\n\n\n\n

    Custom GPTs, Skills, Apps\/Connectors.<\/strong> Custom GPTs = repeatable assistants with instructions + knowledge files. Skills = reusable workflow definitions. Apps\/connectors (60+: Drive, Slack, GitHub, Salesforce\u2026) plus custom MCP servers pipe your real work context in \u2014 and MCP feeds Deep Research too.<\/p>\n\n\n\n

    Atlas & ChatGPT Work.<\/strong> Atlas puts ChatGPT inside the browser; ChatGPT Work (all plans on desktop) is the task-oriented workspace where Free\/Go users can even reach GPT-5.6 Terra.<\/p>\n\n\n\n

    2.3 ChatGPT prompting mastery<\/h2>\n\n\n\n

    The universal five-part skeleton \u2014 role, goal, context, constraints, format:<\/p>\n\n\n\n

    You are [role]. I need [specific goal].\nContext: [audience, situation, what's been tried].\nConstraints: [length, tone, exclusions, must-includes].\nFormat: [structure of the output].\n<\/code><\/pre>\n\n\n\n
    The pattern in action (from the current power-user canon): “Write a 100-word email to B2B SaaS CTOs who attended our demo last week; goal is a scheduled follow-up call; peer-to-peer tone; no sales language; end with one easy question.”<\/em> Specific inputs \u2192 usable output on the first pass.<\/p>\n\n\n\n
    Prefixes that consistently raise quality:<\/p>\n\n\n\n
      \n
    • “Ask me clarifying questions before answering”<\/strong>\u00a0\u2014 the single best prefix for complex requests.<\/li>\n\n\n\n
    • “Think step by step, then give the final answer”<\/strong>\u00a0\u2014 still effective on non-reasoning models; on reasoning models, raise the effort setting instead.<\/li>\n\n\n\n
    • “Here’s an example of what good looks like: [\u2026]”<\/strong>\u00a0\u2014 few-shot beats description.<\/li>\n\n\n\n
    • “Act as a harsh critic of the draft above”<\/strong>\u00a0\u2014 second-pass self-review.<\/li>\n\n\n\n
    • For Deep Research: state scope, sources to prefer\/avoid, time bounds, and the exact deliverable structure \u2014 then\u00a0edit the proposed plan<\/strong>\u00a0before launching; it’s your one cheap steering moment.<\/li>\n\n\n\n
    • For Agent Mode: give the end state and guardrails (“stop and ask before any purchase\/login”).<\/li>\n<\/ul>\n\n\n\n
      \n\n\n\n
      PART 3 \u2014 CODEX: THE COMPLETE PICTURE<\/h1>\n\n\n\n
      3.1 What Codex is now<\/h2>\n\n\n\n
      Forget the 2021 autocomplete API. Codex 2026 is OpenAI’s unified agent system<\/strong> \u2014 an agent loop + shared harness serving every surface, used by ~4M developers weekly. It clones or opens your code, plans, edits across files, runs commands and tests in sandboxes, iterates until green, and opens PRs. The April 16, 2026 “Codex for (almost) everything”<\/strong> update pushed the desktop app beyond code: background Computer Use on Mac<\/strong> (its own cursor while you keep working), an in-app browser<\/strong>, image generation<\/strong>, memory<\/strong>, and background Automations<\/strong> \u2014 positioning Codex as OpenAI’s agentic workspace, not just a coding tool.<\/p>\n\n\n\n
      3.2 Every surface<\/h2>\n\n\n\n
      Surface<\/th>What it’s for<\/th><\/tr><\/thead>
      CLI<\/strong> (codex<\/code>)<\/td>Full-screen terminal UI; the power surface. Open source.<\/td><\/tr>
      Desktop App<\/strong><\/td>Threads, worktrees, visual review, local environments, in-app browser, Chrome extension, Computer Use, Automations, Windows support.<\/td><\/tr>
      IDE Extension<\/strong><\/td>VS Code-class inline workflow with its own slash commands.<\/td><\/tr>
      Web \/ Codex Cloud<\/strong><\/td>Asynchronous cloud tasks in network-isolated sandboxes: queue many in parallel, each returns a PR. Best-of-N with --attempts 1\u20134<\/code>.<\/td><\/tr>
      ChatGPT \/ Mobile<\/strong><\/td>Codex inside ChatGPT; Codex Mobile free on all plans since May 2026.<\/td><\/tr>
      GitHub \/ Slack \/ Linear<\/strong><\/td>Tag Codex on issues\/PRs\/messages; GitHub Action for CI.<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
      All surfaces share auth (ChatGPT login or OPENAI_API_KEY<\/code>), config, and the AGENTS.md convention.<\/p>\n\n\n\n
      3.3 Getting started<\/h2>\n\n\n\n
      npm i -g @openai\/codex        # or brew install codex<\/em>\ncodex login                    # browser OAuth (or device-code flow)<\/em>\ncd your-project && codex       # interactive TUI<\/em>\ncodex \"Explain this codebase to me\"\n<\/code><\/pre>\n\n\n\n
      API-key auth (OPENAI_API_KEY<\/code>) works without OAuth \u2014 ideal for CI \u2014 though some cloud-thread features need the ChatGPT login. Config lives in ~\/.codex\/config.toml<\/code> (profiles supported: codex --profile ci exec \"\u2026\"<\/code>).<\/p>\n\n\n\n
      3.4 Approval modes & sandboxing (know these cold)<\/h2>\n\n\n\n
        \n
      • Auto<\/strong>\u00a0(default): read\/edit\/run freely\u00a0inside the working directory<\/em>; asks before going outside it or touching the network.<\/li>\n\n\n\n
      • Read-only<\/strong>: consultative; browses but changes nothing until you approve a plan.<\/li>\n\n\n\n
      • Full Access<\/strong>\u00a0(--yolo<\/code>): whole machine + network, no asking. Use rarely, only on trusted repos.<\/li>\n<\/ul>\n\n\n\n
        Switch with \/permissions<\/code> mid-session. Cloud tasks run in ephemeral, network-isolated sandboxes scoped to your repo. Auto-review<\/strong> adds a safety pass on the agent’s own changes. Web search defaults to a cached OpenAI-maintained index<\/strong> (reduces prompt-injection exposure from live pages); go live with --search<\/code> or web_search = \"live\"<\/code> \u2014 and still treat web content as untrusted.<\/p>\n\n\n\n
        3.5 AGENTS.md \u2014 Codex’s persistent brain<\/h2>\n\n\n\n
        The equivalent of a project constitution: a markdown file Codex reads every session, with a documented hierarchy (global ~\/.codex\/AGENTS.md<\/code> \u2192 repo root \u2192 subdirectory overrides). Fill it with build\/test commands, style rules that differ from defaults, repo etiquette, and gotchas \u2014 and keep it lean; bloat gets ignored. Alongside it: Rules<\/strong> (deterministic constraints), Memories\/Chronicle<\/strong> (Codex’s own accumulated learnings, with memory-mode controls), Skills<\/strong> (on-demand workflows like $imagegen<\/code>), Hooks<\/strong> (scripts on lifecycle events with live activity display), Subagents<\/strong> (parallel workers configured under [agents]<\/code> in config.toml \u2014 powerful, but explicitly documented to consume more tokens than single-agent runs, and only spawned when you ask), Plugins<\/strong> (installable from GitHub\/git URLs via the marketplace), and MCP servers<\/strong> (codex mcp \u2026<\/code>; popular: Context7 docs lookup, Figma, Playwright, Chrome DevTools, GitHub, Sentry \u2014 and Codex can itself run as<\/em> an MCP server inside another agent).<\/p>\n\n\n\n
        3.6 Command & shortcut reference<\/h2>\n\n\n\n
        Slash commands (TUI):<\/strong> \/model<\/code> (switch mid-session) \u00b7 \/permissions<\/code> \u00b7 \/review<\/code> (inline code review presets: against a base branch, uncommitted changes, a specific commit, or custom instructions \u2014 never touches your working tree; great pre-commit check) \u00b7 \/fork<\/code> \u00b7 \/side<\/code> \u00b7 \/theme<\/code> \u00b7 \/clear<\/code> \u00b7 \/copy<\/code> \u00b7 \/status<\/code> \u00b7 \/exit<\/code> \u00b7 custom commands you author for team workflows.<\/p>\n\n\n\n
        Keyboard:<\/strong> @<\/code> fuzzy file search \u00b7 !cmd<\/code> run shell (output enters context) \u00b7 Tab<\/code> queue follow-up input for the next turn (prompts, slash commands, or !<\/code> commands) \u00b7 Enter<\/code> mid-run to inject instructions into the current<\/em> turn \u00b7 Esc Esc<\/code> walk back and fork<\/strong> from a previous message \u00b7 Ctrl+G<\/code> draft in your $VISUAL\/$EDITOR \u00b7 Ctrl+R<\/code> search prompt history \u00b7 Ctrl+O<\/code> copy last output \u00b7 Ctrl+L<\/code> clear screen without clearing the conversation.<\/p>\n\n\n\n
        CLI essentials:<\/strong><\/p>\n\n\n\n
        codex resume [--last | --all | <SESSION_ID>]   # reopen sessions with full context<\/em>\ncodex exec \"fix the CI failure\"                # non-interactive \/ scripting<\/em>\ncodex exec resume --last \"Implement the plan\"\ncodex cloud                                    # triage cloud tasks from terminal<\/em>\ncodex cloud exec --env ENV_ID --attempts 3 \"\u2026\" # best-of-N cloud runs<\/em>\ncodex -i screenshot.png \"Explain this error\"   # image input (comma-separate multiples)<\/em>\ncodex --cd <path> --add-dir ..\/backend         # working root + extra writable roots<\/em>\ncodex --model gpt-5.5                          # pin model at launch<\/em>\ncodex features list|enable|disable <flag>      # feature flags (e.g. multi_agent)<\/em>\ncodex app-server --listen ws:\/\/\u2026  \/  codex --remote wss:\/\/\u2026   # remote TUI<\/em>\ncodex completion zsh                           # shell completions<\/em>\n<\/code><\/pre>\n\n\n\n
        3.7 What Codex is best at (and worst at)<\/h2>\n\n\n\n
        Sweet spot:<\/strong> asynchronous, bounded, independently testable work \u2014 dependency upgrades, test-coverage expansion, style migrations, doc generation from diffs, bug fixes from issue descriptions \u2014 and queuing many such tasks in parallel sandboxes while you review PRs. Weaker fit:<\/strong> interactive exploratory debugging and tasks dependent on external state, where a local agentic session (Codex CLI in interactive mode \u2014 or, candidly, tools like Claude Code) shines.<\/p>\n\n\n\n
        3.8 Codex prompting patterns that work<\/h2>\n\n\n\n
          \n
        1. Scope like a ticket:<\/strong>\u00a0files involved, expected behavior, how to verify. “Fix the login bug” fails; “Users report login fails after session timeout \u2014 check the token-refresh path, write a failing test that reproduces it, then fix it and run the suite” succeeds.<\/li>\n\n\n\n
        2. Make tests the contract.<\/strong>\u00a0Codex’s lineage is literally\u00a0trained to iterate until tests pass<\/em>\u00a0\u2014 give it tests to pass.<\/li>\n\n\n\n
        3. Pre-set the environment<\/strong>\u00a0(official tip): activate venvs, start daemons, export env vars\u00a0before<\/em>\u00a0launching so Codex doesn’t burn tokens probing.<\/li>\n\n\n\n
        4. Use\u00a0\/review<\/code>\u00a0before every commit<\/strong>\u00a0\u2014 cheap, read-only, prioritized findings; customize (“Focus on accessibility regressions”).<\/li>\n\n\n\n
        5. Fork instead of arguing:<\/strong>\u00a0after two failed corrections,\u00a0Esc Esc<\/code>\u00a0back to the fork point and re-prompt with what you learned.<\/li>\n\n\n\n
        6. Cloud for breadth, CLI for depth:<\/strong>\u00a0batch the bounded stuff to Codex Cloud (--attempts 3<\/code>\u00a0on anything gnarly), keep exploratory work local.<\/li>\n\n\n\n
        7. Attach visuals<\/strong>\u00a0(-i design.png<\/code>) for UI work; ask for a screenshot-vs-spec comparison as verification.<\/li>\n<\/ol>\n\n\n\n
          \n\n\n\n
          PART 4 \u2014 THE COST MASTERY PLAYBOOK<\/h1>\n\n\n\n
          4.1 Subscription-side savings<\/h2>\n\n\n\n
          Buy the cheapest plan that covers your bottleneck feature<\/em>, not the model badge: most knowledge workers cap out at Plus $20 (the honest consensus of every current guide). The $100 Pro exists specifically for people hitting Plus’s Codex\/usage ceilings; the $200 Pro is essentially “the Sora + Operator tier.” Watch limit-based traps: Deep Research (~10\/mo on Plus) and Agent Mode (~40 msgs\/mo) are the quotas real users hit first \u2014 spend them on tasks worth 30 minutes of agent time, use plain Search otherwise. And remember the Auto-router: explicitly selecting the flagship on hard tasks is free quality you’re already paying for.<\/p>\n\n\n\n
          4.2 API-side savings (the big levers)<\/h2>\n\n\n\n
            \n
          1. Model routing:<\/strong>\u00a0Luna ($1\/$6) \u2192 Terra ($2.50\/$15) \u2192 Sol ($5\/$30). Start low, escalate on evidence. Output tokens are 6x input on every tier \u2014 output-heavy work is where tier choice bites.<\/li>\n\n\n\n
          2. Prompt caching: 90% off cached input reads<\/strong>\u00a0(Sol $0.50, Terra $0.25, Luna $0.10 per 1M). Structure prompts with a stable prefix (system prompt, tools, docs) and volatile suffix to maximize hits \u2014 the single biggest lever for agent workloads with big repeated prefixes.<\/li>\n\n\n\n
          3. Batch API: 50% off everything<\/strong>, asynchronous \u2014 for evals, backfills, bulk generation.<\/li>\n\n\n\n
          4. Flex processing:<\/strong>\u00a0discounted, lower-priority compute for latency-tolerant jobs (and Priority processing when latency is the constraint, at a premium).<\/li>\n\n\n\n
          5. Reasoning effort:<\/strong>\u00a0set\u00a0reasoning_effort<\/code>\u00a0low for simple calls \u2014 reasoning tokens bill as output tokens.<\/li>\n\n\n\n
          6. Compaction & conversation state:<\/strong>\u00a0use the Responses API’s context-management features instead of resending ever-growing transcripts.<\/li>\n<\/ol>\n\n\n\n
            4.3 Codex-specific savings<\/h2>\n\n\n\n
            Keep the environment pre-warmed (probing costs tokens). Use subagents only when parallelism genuinely pays \u2014 they’re documented to cost more than single-agent runs. Prefer \/review<\/code> (read-only, one pass) over “re-do it” loops. Cached web search is default and cheaper\/safer than live. Image generation inside Codex burns included limits 3\u20135x faster<\/strong> than normal turns \u2014 for batches, set OPENAI_API_KEY<\/code> and route through the API instead (official recommendation). Fork early rather than accumulating failed-correction context. In cloud, --attempts<\/code> multiplies cost linearly \u2014 reserve best-of-N for genuinely hard tasks.<\/p>\n\n\n\n
            \n\n\n\n
            PART 5 \u2014 CHATGPT vs CODEX: WHICH TOOL FOR WHICH JOB<\/h1>\n\n\n\n
            Job<\/th>Reach for<\/th><\/tr><\/thead>
            Questions, drafting, analysis, brainstorming<\/td>ChatGPT (right model, right effort)<\/td><\/tr>
            A cited multi-source report<\/td>Deep Research (edit the plan first)<\/td><\/tr>
            Long document\/code you’ll iterate on<\/td>Canvas<\/td><\/tr>
            Spreadsheet math<\/td>Data Analysis \/ ChatGPT in Excel<\/td><\/tr>
            A recurring briefing or monitor<\/td>Tasks<\/td><\/tr>
            Multi-step web actions<\/td>Agent Mode \/ Operator<\/td><\/tr>
            Any real coding task<\/td>Codex (CLI\/IDE for interactive, Cloud for bounded batch)<\/td><\/tr>
            Ten independent fixes by Friday<\/td>Codex Cloud, parallel tasks, review PRs<\/td><\/tr>
            Automating your actual desktop apps<\/td>Codex App Computer Use (Mac)<\/td><\/tr>
            Team triage from chat\/issues<\/td>Codex GitHub\/Slack\/Linear integrations<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
            \n\n\n\n
            PART 6 \u2014 SAFETY, GOVERNANCE & FAILURE PATTERNS<\/h1>\n\n\n\n
            Safety:<\/strong> never --yolo<\/code> on untrusted repos; treat web content and issue text as prompt-injection surfaces (cached search default exists for exactly this); review agent transcripts \u2014 Codex always surfaces one, and git is your rollback; Enterprise gets managed configuration, governance, access tokens, and Lockdown Mode; the Codex Security plugin\/cloud adds a dedicated vulnerability-review layer.<\/p>\n\n\n\n
            Failure patterns \u2192 fixes:<\/strong> trusting Auto on hard prompts \u2192 pick the model. Vague tickets \u2192 scope + verification criteria. Correction spirals \u2192 fork (Esc Esc<\/code>) with a better prompt. Burning Deep Research quota on Search-sized questions \u2192 downshift. Memory drift \u2192 review and prune Memory; keep workstream context in Projects. Bloated AGENTS.md \u2192 prune like code. Subagents everywhere \u2192 single agent unless parallelism pays. Unverified research \u2192 click the citations.<\/p>\n\n\n\n
            \n\n\n\n
            PART 7 \u2014 CANONICAL RESOURCES<\/h1>\n\n\n\n
            Codex (official):<\/strong> developers.openai.com\/codex (overview, quickstart, pricing, prompting, models) \u00b7 \/codex\/cli\/features, \/codex\/cli\/reference, \/codex\/cli\/slash-commands \u00b7 \/codex\/guides\/agents-md \u00b7 \/codex\/mcp, \/codex\/skills, \/codex\/subagents, \/codex\/hooks \u00b7 \/codex\/cloud \u00b7 \/codex\/learn\/best-practices \u00b7 \/codex\/changelog \u00b7 github.com\/openai\/codex (open source, releases) ChatGPT:<\/strong> help.openai.com (release notes, feature articles) \u00b7 openai.com\/chatgpt\/pricing \u00b7 chatgpt.com API:<\/strong> developers.openai.com\/api\/docs \u2014 prompting & prompt-engineering guides, reasoning best practices, prompt caching, batch, flex processing, cost & latency optimization \u00b7 platform.openai.com (dashboard) \u00b7 openai.com\/api\/pricing<\/p>\n\n\n\n
            \n\n\n\n
            Compiled July 28, 2026. Fastest-moving details \u2014 GPT-5.6 rollout by plan\/region, Deep Research and Agent Mode quotas, Task limits (OpenAI’s own docs are currently inconsistent above Plus), and Codex CLI versions (shipping near-weekly) \u2014 deserve a fresh check at the official links above before you commit budget.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"
            Gold & Diamond Standard Edition \u2014 Verified against live sources, July 28, 2026 How this guide was built: Compiled on July […]<\/p>\n","protected":false},"author":4,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-3868","post","type-post","status-publish","format-standard","hentry","category-uncategorized"],"_links":{"self":[{"href":"http:\/\/aiopsschool.com\/blog\/wp-json\/wp\/v2\/posts\/3868","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/aiopsschool.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/aiopsschool.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/aiopsschool.com\/blog\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"http:\/\/aiopsschool.com\/blog\/wp-json\/wp\/v2\/comments?post=3868"}],"version-history":[{"count":1,"href":"http:\/\/aiopsschool.com\/blog\/wp-json\/wp\/v2\/posts\/3868\/revisions"}],"predecessor-version":[{"id":3869,"href":"http:\/\/aiopsschool.com\/blog\/wp-json\/wp\/v2\/posts\/3868\/revisions\/3869"}],"wp:attachment":[{"href":"http:\/\/aiopsschool.com\/blog\/wp-json\/wp\/v2\/media?parent=3868"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/aiopsschool.com\/blog\/wp-json\/wp\/v2\/categories?post=3868"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/aiopsschool.com\/blog\/wp-json\/wp\/v2\/tags?post=3868"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}