Tools

Interactive

What will the cloud bill be?

Start from a typical setup and change the sizes. You get a monthly estimate at Azure list prices, split by what drives it, and the lever for the biggest part.

• Compute$1,679 · 53%
• Logs$909 · 29%
• Database$310 · 10%
• Data transfer$165 · 5%
• Disks$108 · 3%
• Object storage$9.80 · 0%

Azure West Europe, pay-as-you-go list prices in USD as of 8 November 2026 (Azure Retail Prices API): compute at D4s v5 Linux, disks at Premium SSD P10, database at PostgreSQL flexible server General Purpose (compute only), data transfer via the Microsoft network after 100 GB free, logs at Log Analytics Analytics logs. Not included: reservations, licences, backup, support, taxes.

Azure Retail Prices API

Interactive

What does an AI feature cost per month?

Put in your traffic and prompt sizes. You get the monthly bill on your model, on all current models, and what prompt caching and routing simple requests to a small model save.

The same traffic on every model (with your caching, without routing)

• Claude Haiku 5.5 Anthropic$27
• GPT-6 Luna OpenAI$27
• Gemini 3.5 Flash-Lite Google$112
• GPT-6.1 Terra OpenAI$274
• Gemini 3.8 Flash Google$410
• Claude Sonnet 5.5 Anthropic$547
• GPT-6.1 Sol OpenAI$547
• Claude Opus 5.5 Anthropic$1,094
• Gemini 4 Argon Google$1,094
• Claude Fable 5.1 Anthropic$2,736
• GPT-6 Astra OpenAI$2,736

Standard list prices per million tokens as of 8 November 2026. Not included: cache writes, long-context surcharges, tool and search fees, taxes. Gemini 4 Argon and 3.8 Flash show their regular prices; introductory discounts run until 30 November and 31 December 2026. Prices change: check the pricing pages before you budget.

Anthropic · OpenAI · Google

Interactive

Does the commitment pay off?

A reservation or savings plan is cheaper per hour, but you pay for it whether you use it or not. Here is the utilisation from which it pays off, and what is left over the term.

My rule: buy after right-sizing, and only for the base that has run steadily for months. Below break-even you pay for capacity you don't use.

Interactive

What do your logs cost?

Log volume and retention, priced on Azure, AWS and Google Cloud. Most of the time the bill comes from what you write, not from how long you keep it.

Writing is 87–95% of the cost here: logging less (debug levels, health checks, noisy services) saves more than shorter retention.

List prices in USD as of 8 November 2026: Azure Log Analytics (Analytics logs, West Europe, 31 days included), AWS CloudWatch Logs (Standard, Frankfurt; storage billed on compressed size, so this is an upper bound), Google Cloud Logging (30 days included, first 50 GiB per project and month free; GiB counted as GB). Azure and AWS free allowances, cheaper log tiers and archive options are not included.

Azure Retail Prices API · Azure Monitor Logs · AWS CloudWatch · Google Cloud

Interactive

How many nodes do your requests really need?

Move the sliders to your numbers. Requests decide where pods fit, so oversized requests buy nodes nobody uses.

CPU only, about 90% of a node is allocatable, at least 2 nodes for availability. Check memory the same way before you remove a node.

Interactive

Which device for a local LLM?

Pick a model and how you will use it. You get the memory it needs and the smallest machines that hold it, from a mini PC to a GPU server.

All options, roughly from small to large investment

• PC with GeForce RTX 5090Too small

  Fast for models that fit into 32 GB. A consumer card: loud, hot, no ECC memory, 575 W total graphics power.

• Mini PC with AMD Ryzen AI Max+ 395Fits · 128 GB

  The cheapest route to a lot of memory: up to 128 GB of 256-bit LPDDR5x-8000 (about 256 GB/s) shared with the Radeon 8060S graphics, 45 to 120 W. Runs llama.cpp, LM Studio and Ollama.

  ≤ 7 tokens/s, upper limit for one user on that machine

• NVIDIA DGX SparkFits · 128 GB

  GB10 superchip with 128 GB unified memory and the full CUDA stack. 273 GB/s limits large dense models; strong for mixture-of-experts models, development and fine-tuning experiments.

  ≤ 8 tokens/s, upper limit for one user on that machine

• Mac Studio M5 MaxFits · 128 GB

  Up to 128 GB unified memory at 460 to 614 GB/s, and quiet. MLX, llama.cpp, Ollama and LM Studio run well on it.

  ≤ 13–18 tokens/s, upper limit for one user on that machine

• Workstation or server with RTX PRO 6000 Blackwell (1–2)Fits · 96 GB

  96 GB ECC GDDR7 at 1,792 GB/s and 600 W per card: the fastest per user. One or two of them with vLLM make a server for a team.

  ≤ 51 tokens/s, upper limit for one user on that machine

• Rack server with 2× L40SFits · 96 GB

  Two 48 GB data-center GPUs with vLLM in front: the other team setup from the article for 30B to 70B models.

• Mac Studio M5 UltraFits · 512 GB

  Up to 512 GB unified memory at 1.2 TB/s, at most 480 W continuous: the simplest way to fit a very large model on a desk.

  ≤ 34 tokens/s, upper limit for one user on that machine

• NVIDIA DGX Station (GB300)Fits · 748 GB

  A deskside machine with the GB300 superchip and 748 GB of coherent memory, 252 GB of it HBM3e.

• GPU server with H200 (up to 8)Fits · 141 GB

  141 GB HBM3e at 4.8 TB/s and up to 700 W per GPU; eight of them give over a terabyte. vLLM splits the model across the GPUs.

• NVIDIA DGX B200Fits · 1,440 GB

  Eight B200 GPUs with 1,440 GB HBM3e and 64 TB/s in total, at about 14.3 kW for the system. Power, cooling and operations become the bigger project.

Planning estimate, not a benchmark. Memory = parameters × bytes per parameter, plus about 20% for the KV cache (long documents or a few people: +50%, a team: +100%, rough factors). Unified-memory machines keep about 10% for the system. The speed limit is memory bandwidth divided by the weights read per token (only the active ones for mixture-of-experts); real numbers are lower. Device figures from the vendors' spec pages, as listed in the article (Ryzen bandwidth calculated from AMD's 256-bit LPDDR5x-8000). Before buying, test your model and prompts on rented GPUs for two weeks.

Interactive

Which GPU does your model need?

Pick a model size and precision. Weights take parameters × bytes per parameter; about 20% on top covers the KV cache for normal context lengths.

• 1× 24 GB (RTX 4090)
• 1× 32 GB (RTX 5090)
• 1× 48 GB (L40S)
• 1× 80 GB (H100)
• 1× 96 GB (RTX PRO 6000 Blackwell)
• 1× 141 GB (H200)

Rule of thumb for planning, not a benchmark. Long contexts and many parallel users need more KV cache; 4-bit costs a little quality.

Interactive

What does a mixture-of-experts model cost to host?

All experts must sit in GPU memory, but each token only runs through the active ones. Memory scales with total parameters, speed with active parameters.

Planning estimate: weights plus about 15% for KV cache and runtime. Real throughput depends on batch size, interconnect and the serving engine.

Interactive

Is your air gap more than a firewall rule?

Tick what your enclave already has. The gaps are where audit findings and outages usually start.

• A written threat model: what must never leave, what must never come in
• One defined transfer path with scan, checksum and signature check
• A private registry mirror that only accepts signed images
• An SBOM for everything that is imported
• GitOps from a Git server inside the enclave
• A time source inside the enclave
• Monitoring, logs and alerts stay inside
• A fixed patch cadence with tested update bundles
• Break-glass access documented and rehearsed
• A restore has been tested inside the enclave

Interactive

Is your AI-built app ready for production?

Tick what is already true. The gaps that are left are your to-do list, in order.

• Code lives in Git; changes go through pull requests
• No secret keys in the frontend; server keys only on the server
• Row-level security (or equivalent) on every table
• Database changes as versioned migrations
• A separate staging environment with its own data
• Automated tests run before every merge
• Smoke test after every deploy, and a one-click way back
• Encrypted backups outside the hosting provider
• A restore has actually been tested
• Uptime and business checks that alert a person
• Budget alerts on the cloud and AI bills
• Privacy basics: data map, deletion on request, processor contracts

Interactive

Which mode fits your task?

Answer three questions about the task you want the agent to take off your hands.

1. 1.Does it have a finish line a machine can check (tests green, file exists, metric reached)?

   YesNo

2. 2.Should it repeat on an interval or keep watching something?

   YesNo

3. 3.Must it run while your laptop is off, or start from a schedule, an API call or a GitHub event?

   YesNo

Interactive

Which Codex settings fit your task?

Pick the situation. You get the sandbox mode, the approval policy and the config to copy, using the setting names from the Codex docs (November 2026).

The docs' lower-risk automation preset: edits inside the workspace, asks before anything else.

# ~/.codex/config.toml
sandbox_mode = "workspace-write"
approval_policy = "on-request"

[sandbox_workspace_write]
network_access = false

Interactive

Compare the three side by side

Pick what matters most to you. Facts as of November 2026, from the vendors' docs and changelogs listed under sources.

Interactive

Which Claude plan fits how you work?

Pick the description closest to you. Plans and prices as of the change on 30 October 2026 (USD), from Anthropic's pricing page and help center.

Interactive

How ready is your brand for AI answers?

Tick what is already true. What's left is your GEO to-do list, most important first.

• Brand facts are identical on the site, Google Business Profile, LinkedIn and directories
• Key pages answer the main question in the first paragraph
• Organization or LocalBusiness structured data on key pages
• A written robots.txt decision for GPTBot, PerplexityBot and Google-Extended
• Key content is in the HTML, not only after JavaScript runs
• FAQ and comparison pages for what customers really ask
• Listings and reviews maintained on the portals of your industry
• Original data or expertise others can quote
• AI-referred visits tracked separately in analytics
• Share of AI answers tracked for your top questions (Semrush, Peec AI)

Interactive

What do your customers ask an AI?

Pick an industry and the intents you care about. These examples come from templates in your browser; real prompt research generates hundreds per page with an open model on your sitemap.

1. What is the best restaurant in Munich?
2. Which restaurant in Munich do locals recommend?
3. Compare the top three restaurant options in Munich by price and reviews
4. Your brand or another restaurant in Munich: which is better for a small team?
5. Where can I book a restaurant in Munich for this weekend?
6. Find a restaurant in Munich with availability tomorrow