AI Guide

Q2

What matters most?

Q2

How do you code with AI?

Q3

Budget and intensity?

Q2

What is most important?

Q2

What kind of research?

Q2

Which criterion?

RESULT

Study, essays, PDFs, images, everyday use

ChatGPT Plus$20/mo. For heavy use, ChatGPT Pro ($100 or $200/mo — the $200 tier has the higher allowances).

Models

GPT-5.5 / Claude Sonnet 5 / Gemini 3.1 Pro

Why

Strong at explanations, PDFs, math, and structure. ChatGPT covers images, files, code, and search in one plan; Codex is included.

API?

Not needed for studying — the subscription is enough. Building something? Re-run with that answer.

RESULT

Terminal coding agent, best quality, daily heavy use

Claude Max5x from $100/mo, 20x from $200/mo. Note: claude.ai, Claude Code, and Claude Desktop share one usage pool.

Models

Claude Sonnet 5 / Opus 4.8 / Kimi K2.7 Code / DeepSeek V4 Pro

Alternatives

IDE-first? Cursor Pro. GitHub-centric? Copilot Pro+ ($39/mo, AI-credits system).

Cheaper?

See the “pay-as-you-go” and “flat coding plan” answers in this quiz.

RESULT

Coding inside the IDE / editor

Cursor ProFrom $20/mo (Pro / Pro+ / Ultra). GitHub-centric? GitHub Copilot Pro $10 or Pro+ $39/mo.

Notes

Cursor plans include model usage; overages billed on-demand. Copilot uses AI Credits ($0.01/credit; Pro $15, Pro+ $70, Max $200 worth included monthly) — completions stay unlimited.

Caution

Neither is an API provider — you cannot power your own app with them.

RESULT

AI in my own app — production stability first

An API provider — never a personal subscriptionOpenAI API / Anthropic API / Gemini API (+ Vertex AI, Azure, Bedrock)

Production picks

OpenAI GPT-5.6 family (sol $5/$30, terra $2.50/$15, luna $1/$6 per 1M tokens), Anthropic Sonnet 5 ($2/$10 intro until 2026-08-31, then $3/$15), Gemini 3.1 Pro ($2/$12 up to 200k).

Cost picks

DeepSeek V4 Flash $0.14/$0.28, MiniMax M2.7 $0.30/$1.20, Z.ai GLM-5.2 $1.4/$4.4, Kimi K2.7 Code $0.95/$4 — cheap enough to run agents at volume.

Caution

Coding-plan API keys (Alibaba etc.) explicitly prohibit app backends and automation.

RESULT

Start free — or under $10/month

Groq + Cerebras + Gemini APIAll three have real free tiers. Add OpenRouter free models and Z.ai's free GLM Flash models for breadth.

Details

Groq and Cerebras: free API keys, blazing speed. Gemini API: free tier across models. Z.ai: GLM-4.7-Flash / 4.5-Flash are $0. OpenRouter: 50 free-model requests/day (1,000/day once you have bought $10 of credits).

RESULT

Speed above everything — low-latency UI, bulk processing

Groq / CerebrasPurpose-built fast inference, both with free tiers. Cerebras Developer: $10 buys ~10x rate limits.

Models

Groq GPT-OSS 20B ($0.075/$0.30) / 120B ($0.15/$0.60) / Gemini 3.1 Flash-Lite ($0.25/$1.50) / DeepSeek V4 Flash ($0.14/$0.28) / GLM Flash (free)

Why

Cheap, fast, and good enough for sub-agents, classification, and short answers.

RESULT

Web research with citations — reading or building

Perplexity Pro$20/mo; Max $200/mo adds 10,000 monthly credits for the multi-step Computer agent. Ask searches don't consume credits.

Build with API

Perplexity Sonar $1/$1 (Pro $3/$15) + search fees $5–14 per 1k requests / Gemini Grounding / Groq Compound (search $5–8 per 1k) / OpenRouter Web Search.

Caution

The Perplexity subscription does NOT include API usage — the API is separate pay-as-you-go.

RESULT

Google integration — Drive, NotebookLM, storage

Google AI Pro€21.99/mo (German pricing; varies by country). 5 TB storage, full Gemini 3.1 Pro with 1M context, NotebookLM, Flow, Jules.

Tiers

AI Plus €4.99 (400 GB, 2x limits, Gemini 3.5 Flash + limited 3.1 Pro) / AI Pro €21.99 (5 TB, 4x) / AI Ultra €99.99 (5x) or €219.99 (20x, 20 TB+, Deep Think, Spark).

Caution

The Gemini API is a separate product — the subscription does not include API quota.

RESULT

Flat, cheap monthly coding plans

Kimi Code / Z.ai / MiniMaxKimi Code from $15/mo (annual) · Z.ai GLM Coding Plan from $18/mo · MiniMax Token Plan from $20/mo · Alibaba Coding Plan $50/mo.

Quotas

Kimi: weekly-refresh quota (K2.7 Code). Z.ai: ~80/400/1,600 prompts per 5h (Lite/Pro/Max). MiniMax: ~3–7 concurrent agents, 5h + weekly windows. Alibaba: 6,000 req/5h, 45k/week, 90k/month.

Caution

These keys are for interactive coding tools (Claude Code, Cline, OpenCode…). Alibaba explicitly prohibits app backends and automated scripts — check each plan's terms.

RESULT

Claude Code / OpenCode / Cline — pay-as-you-go, cheap

Anthropic API / DeepSeek / OpenRouterSonnet 5 at intro $2/$10 per 1M tokens (until 2026-08-31) is the reference. DeepSeek's Anthropic-compatible endpoint runs Claude Code at V4 prices.

Details

DeepSeek V4 Flash $0.14/$0.28, V4 Pro $0.435/$0.87 via api.deepseek.com/anthropic (official Claude Code guide). OpenRouter: many models, no markup, 5.5% credit fee. Prompt caching cuts input costs further.

For Cline/OpenCode

OpenAI-compatible APIs are easiest: OpenRouter, Groq, DeepSeek, Kimi, MiniMax, Z.ai, Gemini API.

Flat-rate?

Prefer a flat monthly plan? See the coding-plan result.

RESULT

Many models behind one API key

OpenRouterGPT, Claude, Gemini, DeepSeek, Kimi, Qwen… unified billing, no markup on inference; 5.5% fee when buying credits (5% via crypto).

Free tier

Free models: 50 requests/day under $10 of purchased credits, 1,000/day at $10+.

Caution

Always check the underlying provider's terms, logging, region, and price before production.

RESULT

Research PDFs, codebases, very long chats

The 1M-context classGemini 3.1 Pro / Claude Sonnet 5 / DeepSeek V4 / Grok 4.3

Verified

All four offer 1M-token context at standard pricing (Gemini 3.1 Pro, Anthropic Sonnet 5 & Fable 5, DeepSeek V4 Flash/Pro, xAI Grok 4.3). Kimi K2.7 Code is 262k.

Subscription?

Google AI Pro gives Gemini 3.1 Pro with 1M context in the Gemini app; NotebookLM is ideal for a fixed set of documents.

RESULT

Images, video, audio — multimodal in and out

Gemini / ChatGPTBroadest multimodal coverage. Google AI adds Flow (video credits by tier); MiniMax covers text, image, speech, and music on one platform.

Notes

For API work, Gemini API's multimodal endpoints are the easiest starting point; Mistral offers OCR for documents.

RESULT

X (Twitter) and real-time information

Grok / xAI APIAPI verified: grok-4.3 $1.25/$2.50 per 1M with 1M context; grok-4.5 $2/$6 (500k). Consumer SuperGrok pricing couldn't be verified — check the official site.

Why

Best affinity with X and real-time data; grok-build-0.1 ($1/$2, 256k) targets coding.

RESULT

A European provider

Mistral Pro$14.99/mo (Education $5.99, Team $24.99/user). API: Mistral Large $2/$6 per 1M, Batch 50% off, OCR available.

Why

European company, clean UI, coding via Le Chat (Vibe); enterprise-friendly API with document/OCR strengths. Le Chat and the API are billed separately.

Free $0 / Go $8 / Plus $20 / Pro $100 · $200 / Business · Enterprise

Models

GPT-5.5 and GPT-5.4 on all paid plans; Plus and up add GPT-5.2 Thinking and Codex. (GPT-5.5 Pro is API-only.)

Usage

Grows Go → Plus → Pro. Pro has two tiers: $100 and $200; $200 carries the higher allowances.

API

The OpenAI API is sold separately. Codex is included in plans (its API priced separately).

Verdict

The all-rounder: study, essays, images, code, and search in one plan.

Free $0 / Pro $20 ($17 annual) / Max 5x $100+ / Max 20x $200+ / Team $20/seat

Models

Fable · Opus · Sonnet · Haiku

Usage

Max gives 5x or 20x the Pro usage. claude.ai, Claude Code, and Claude Desktop share one usage pool.

API

The Anthropic API is sold separately.

Verdict

Top of the class for coding agents. Heavy daily Claude Code use points to Max.

German pricing (varies by country): AI Plus €4.99 / AI Pro €21.99 / AI Ultra €99.99 (5x) · €219.99 (20x)

Models

Plus: Gemini 3.5 Flash + limited 3.1 Pro. Pro: full Gemini 3.1 Pro (1M context). Ultra: Deep Think & Gemini Spark early access. Plus NotebookLM, Flow, Antigravity, Jules.

Usage

Plus 2x / Pro 4x vs free; Ultra up to 20x of Pro. Storage: 400 GB / 5 TB / 20 TB+.

API

The Gemini API is a separate product with separate billing.

Verdict

Unbeatable if you live in Drive, NotebookLM, and need storage anyway.

Free $0 / Pro $20 / Max $200 ($2,000/yr)

Models

Selectable inside Perplexity: GPT-5.2, Claude Sonnet 4.6, Gemini 3.1 Pro, NVIDIA Nemotron and more.

Usage

Max: 10,000 credits/mo for the multi-step Computer agent; Ask searches are credit-free. Unused monthly credits expire.

API

The Sonar API is separate pay-as-you-go — not included in Pro/Max.

Verdict

The strongest tool for cited web research; not a coding-agent product.

Hobby Free / Individual from $20 (Pro · Pro+ · Ultra) / Teams $40/user

Models

“Frontier models” (Claude / GPT / Gemini class) inside the Cursor editor.

Usage

Each plan includes a set amount of model usage; going over is billed on-demand in arrears.

API

Not an API provider — cannot power your own app.

Verdict

The strongest choice if you want everything inside one IDE.

Free $0 / Pro $10 / Pro+ $39 / Max $100 (+ Business & Enterprise)

Models

GPT, Claude (Haiku/Sonnet/Opus), and Gemini variants; Copilot coding agent; Claude Code / Codex integrations.

Usage

AI Credits at $0.01/credit — roughly $15 (Pro), $70 (Pro+), $200 (Max) worth included monthly. Code completions are unlimited and free of credits.

API

Not an API provider.

Verdict

The stable pick if your life is on GitHub.

Free $0 / Pro $14.99 / Team $24.99/user / Education $5.99

Models

Mistral Large, Medium 3.5, Small, OCR — through Le Chat (Vibe) for chat and coding.

Usage

Pro raises message, search, image, and Vibe coding limits over Free.

API

Billed separately from Le Chat.

Verdict

The European choice, with a genuinely nice UI.

Annual billing, per month: Moderato $15 / Allegretto $31 / Allegro $79 / Vivace $159

Models

Kimi K2.7 Code

Usage

Weekly-refreshed quotas; higher tiers add quota and concurrency. For terminal and IDE plugins.

API

The Kimi/Moonshot API is separate.

Verdict

Strong at coding and long text for the price.

Lite from $18/mo (Pro/Max exist; prices not published on the official docs)

Models

GLM-5.2 · GLM-5-Turbo · GLM-4.7

Usage

About 80 / 400 / 1,600 prompts per 5-hour window (Lite / Pro / Max).

API

Aimed at Claude Code, Cline, OpenCode and similar tools.

Verdict

One of the cheapest usable coding plans; not for public app backends.

Plus $20 / Max $50 / Ultra $120

Models

All models on the MiniMax API Platform — text, image, speech, and music share one quota.

Usage

Roughly 3–4 / 4–5 / 6–7 concurrent agents; 5-hour rolling plus weekly windows.

API

Subscription-key style, tied to the API platform.

Verdict

Remarkably usable at $20 for agent work.

Pro $50/mo — the only tier (Lite closed to new subscriptions since 2026-03)

Models

qwen3.7-plus · qwen3.6-plus · kimi-k2.5 · glm-5 · MiniMax-M2.5 …

Usage

6,000 req/5h · 45,000/week · 90,000/month

API

Plan API key is for interactive coding tools only (Claude Code, OpenCode, Cursor, Cline…). Backends and automated scripts are explicitly prohibited — violations can suspend the subscription.

Verdict

Great request volume for the price — if you respect the strict terms.

$4.99 / $19.99 / $49.99 / $99.99 / $249.99 per month (points)

Models

Many: GPT-5.5, Claude Opus 4.7, Gemini 3.5 Flash, Veo 3.1, Sora 2 and more.

Usage

Cheapest tier: 10,000 points/day; higher tiers 660k–8.25M points/month; add-on points available.

API

A Poe API exists; web and API draw from the same points.

Verdict

Low priority in this guide — for the same goal, OpenRouter + free tiers usually serve better.

Free tier / pay-as-you-go / Batch API −50%

Models

GPT-OSS 20B $0.075/$0.30 · GPT-OSS 120B $0.15/$0.60 · Llama 3.3 70B $0.59/$0.79 · Qwen3 32B $0.29/$0.59 · Compound

Use for

Apps, CLIs, agents; OpenAI-compatible endpoint.

Verdict

Extreme speed. Compound bundles web search ($5–8/1k) and code execution ($0.18/hr) as built-in tools.

Free tier / Developer from $10 (≈10x rate limits) / Cerebras Code subscription

Models

GPT-OSS 120B · Gemma 4 31B (preview) · Z.ai GLM 4.7 (preview)

Use for

Fast inference, prototypes, agents.

Verdict

Groq's rival for raw speed. Model catalog is small and changes — check before committing.

No markup on inference; 5.5% fee when buying credits ($0.80 min; 5% via crypto)

Models

OpenAI, Anthropic, Google, DeepSeek, Kimi, Qwen and hundreds more — at each provider's own rate.

Free tier

Free models: 50 requests/day (1,000/day once you've bought $10+ of credits).

Verdict

The router: one key, unified billing, drop-in OpenAI-compatible. Verify the underlying provider's terms for production.

Free tier / paid per model

Models

Gemini 3.1 Pro $2/$12 (≤200k; $4/$18 beyond) · 3.5 Flash $1.50/$9 · 3.1 Flash-Lite $0.25/$1.50 · 3 Flash Preview $0.50/$3

Use for

Multimodal, 1M-token long context, Grounding (web search).

Verdict

Separate from the Google AI subscription. The free tier is a real on-ramp; Flash-Lite is a bargain workhorse.

Pay-as-you-go; batch/flex tiers with discounts

Models

GPT-5.6-sol $5/$30 · 5.6-terra $2.50/$15 · 5.6-luna $1/$6 · gpt-5.4-mini $0.75/$4.50 · nano $0.20/$1.25 · gpt-5.5-pro $30/$180

Use for

Production apps, agents, images, audio, Codex-style coding.

Verdict

The GPT-5.6 family (sol/terra/luna) is the current flagship line. Stable and complete — not the cheapest.

Sonnet 5 intro $2/$10 until 2026-08-31, then $3/$15 · Opus 4.8 $5/$25 · Fable 5 $10/$50 · Haiku 4.5 $1/$5

Context

1M tokens at standard pricing on Fable 5, Opus 4.6–4.8, Sonnet 5 & 4.6.

Use for

Coding, long documents, agents, document analysis.

Verdict

Claude Code's home. Prompt caching (0.1x reads) and Batch −50% cut real costs a lot.

V4 Flash $0.14/$0.28 · V4 Pro $0.435/$0.87 (cache-hit input ≈$0.003–0.004)

Context

1M tokens on both; up to 384k output.

Use for

Coding, agents, low-cost backends.

Verdict

OpenAI-compatible by default plus an official Anthropic-compatible endpoint (api.deepseek.com/anthropic) documented for Claude Code. Note: a China-based provider — weigh data routing for your use.

K2.7 Code $0.95 in / $0.19 cached / $4 out · highspeed $1.90 / $0.38 / $8

Context

262k tokens; caching cuts input cost 80%.

Use for

Coding, long text, agents; OpenAI-compatible.

Verdict

Coding quality that rivals far pricier models. China-based provider — weigh data routing.

M2.7 $0.30/$1.20 · M3 $0.30/$1.20 (≤512k; $0.60/$2.40 beyond) · highspeed/Priority cost more

Use for

Agents, coding, multimodal (text, image, speech, music on one platform).

Verdict

M3's price reflects a “permanent 50% off”. Pairs with the $20 Token Plan for solo devs. China-based provider.

GLM-5.2 $1.4/$4.4 · GLM-4.7-Flash & GLM-4.5-Flash free

Use for

Coding, low-cost APIs, web search tool at $0.01/use.

Verdict

The free Flash models are genuinely useful for sub-agents and drafts. China-based provider.

Sonar $1/$1 · Sonar Pro $3/$15 · Reasoning Pro $2/$8 · Deep Research $2/$8 (+fees)

Search fees

Per 1k requests: Sonar $5–12, Sonar Pro / Reasoning Pro $6–14 (by search-context size); Deep Research queries $5/1k.

Use for

Web search with citations, research features inside your app.

Verdict

The reference web-search API. Completely separate billing from Pro/Max subscriptions.

grok-4.3 $1.25/$2.50 (1M ctx) · grok-4.5 $2/$6 (500k) · grok-build-0.1 $1/$2 (256k)

Use for

X/real-time information, long context, coding (grok-build).

Verdict

Strong X affinity at aggressive prices. Consumer SuperGrok pricing was not verifiable on official pages (2026-07-09) — check x.ai.

Mistral Large $2/$6 · Batch −50% · OCR available

Use for

EU-based API, documents, OCR, agents.

Verdict

The European pick for enterprises and document pipelines. Billed separately from Le Chat.