Quick reference for picking a provider/CLI on day 1. All numbers verified early-mid 2026 — confirm live before you depend on them. Pricing and quotas change frequently (especially Chinese providers and free-tier limits).
| Goal | Pick | Why |
|---|---|---|
| Zero cost, zero card | ||
| Gemini CLI with personal Google account | ||
| Generous free tier, no card required, 60 req/min | ||
| Zero cost + multi-provider | opencode + Gemini API key (AI Studio free tier) | |
| Model-agnostic CLI; swap providers later without changing tool | ||
| Best quality, paid | Claude Code with Anthropic API or Bedrock | Top coding quality; budget $5–10/student/month is realistic |
| Cheapest paid bulk | DeepSeek V4 Flash via opencode | $0.14/$0.28 per M tokens — 20× cheaper than Claude |
If your region blocks claude.ai
, skip Claude Code on day 1 — use Gemini CLI or opencode instead. We'll add the regional fallback later.
Provider: Google Gemini Code Assist (consumer tier).Auth: browser sign-in with a personal Google account on first run.Quota: up to60 requests/minute,~1,000 requests/day on Flash; cap drops to ~100/day on Pro-class models.Cost:****$0. No card. May log prompts for training (read the consent screen).** Install:**npm i -g @google/gemini-cli
→gemini
.
Provider:ai.google.dev
— separate from Gemini CLI; gives you an API key you can wire into opencode/LiteLLM/anything.Free quotas(per-day, per-model, no card):** Gemini 2.5 Pro**— ~100 requests/day · 5 req/min** Gemini 2.5 Flash**— ~250 requests/day · 10 req/min** Gemini 2.5 Flash-Lite**— ~1,500 requests/day · 15 req/min
Cost:****$0 within the quota; over-quota = 429 until reset.Get a key:https://aistudio.google.com/app/apikey
**Free BYOK:**1,000,000 requests/month routed through OpenRouter to your own provider keys, then 5% surcharge.real free hosted variants (Qwen3-Coder, DeepSeek, GLM, Kimi) rate-limited to:free
models:1,000 req/day after a one-time $10 top-up;** 50/day without**.** Catch:free traffic may be logged/used for upstream training. Get a key:**https://openrouter.ai/
Numbers below are list pricing per 1,000,000 tokens, input then output. Cached input read = the price for repeated context (huge cost saver for coding agents that resend the same files).
| Provider / Model | Input | Cached read | Output |
|---|---|---|---|
| Anthropic Claude Sonnet 4.5 | |||
| $3.00 | $0.30 | $15.00 | |
| Anthropic Claude Haiku 4.5 | |||
| $1.00 | $0.10 | $5.00 | |
| OpenAI gpt-5-codex | |||
| $1.25 | $0.125 | $10.00 | |
| Google Gemini 2.5 Pro | |||
| $1.25 (≤200k ctx) | $0.31 | $10.00 | |
| Google Gemini 2.5 Flash | |||
| $0.30 | $0.075 | $2.50 | |
| Google Gemini 2.5 Flash-Lite | |||
| $0.10 | low | $0.40 | |
| Z.ai GLM-4.6 | |||
| $0.39 | low | $1.74 | |
| MiniMax M2 | |||
| $0.255 | none | $1.00 | |
| Alibaba Qwen3-Coder | |||
| $0.22 | none | $0.90 | |
| DeepSeek V4 Flash | |||
| $0.14 | $0.0028 | $0.28 | |
| DeepSeek V4 Pro (list) | |||
| $1.74 | $0.0145 | $3.48 | |
| DeepSeek V4 Pro (promo to 2026-05-31) | |||
| $0.435 | $0.0036 | $0.87 |
Sources to verify:
- Anthropic:
[https://www.anthropic.com/api](https://www.anthropic.com/api)(or`platform.claude.com/docs/en/about-claude/pricing`
) - OpenAI:
[https://developers.openai.com/codex/pricing](https://developers.openai.com/codex/pricing) - Gemini:
[https://ai.google.dev/gemini-api/docs/pricing](https://ai.google.dev/gemini-api/docs/pricing) - DeepSeek:
[https://api-docs.deepseek.com/quick_start/pricing](https://api-docs.deepseek.com/quick_start/pricing) - MiniMax:
[https://platform.minimax.io/docs/pricing/overview](https://platform.minimax.io/docs/pricing/overview) - Z.ai / GLM:
[https://docs.z.ai/guides/overview/pricing](https://docs.z.ai/guides/overview/pricing) - OpenRouter:
[https://openrouter.ai/](https://openrouter.ai/)
| Plan | $/month | What you get | Works with |
|---|---|---|---|
| Claude Pro | |||
| $20 | 5h session cap + weekly cap (~40–80 Sonnet hrs/week) | Claude Code (native) | |
| Claude Max 5x / 20x | |||
| $100 / $200 | 5× / 20× Pro | Claude Code (native) | |
| ChatGPT Go / Plus / Pro | |||
| $8 / $20 / $100 | message-based (15–80 msgs / 5h on Plus) | Codex CLI | |
| GitHub Copilot Free / Pro / Pro+ | |||
| $0 / $10 / $39 | usage credits (moving to credits Jun 2026) | Copilot agent | |
| Cursor Pro | |||
| $20 | ~$20 credit pool (~225 Sonnet req/mo) | Cursor IDE | |
| Z.ai GLM Coding Plan Lite / Pro | |||
| $10 / $30 | ~120 prompts/5h (Lite); GLM-5.x models | Claude Code, opencode, Cursor | |
| MiniMax Coding Starter | |||
| $10 | ~100 prompts / 5h on M2.1 | Any | |
| opencode Go | |||
| $10 ($5 first month) | $60/mo usage value across 13 open models | opencode (native) | |
| opencode Zen |
pay-as-you-go | per-token gateway, mostly $0.30–$3.20/M | opencode (native) | Important: quota caps on flat subs are usually measured in prompts (not tokens). If you blow through "120 prompts / 5h" with one heavy session you wait 5h — not great for an active coding day. For predictable cost on a budget, token-metered APIs through one gateway usually win.
Per-prompt is tiny; per-day is what matters. Coding-agent turn averages ~5,000 input + 500 output tokens (lots more if you load a big file). With 60% cache-hit (typical for repeat-edit sessions):
| Provider / Model | Tokens per $1 (light coding) | Approx prompts |
|---|---|---|
| Claude Sonnet 4.5 | ~250k mixed | ~40 prompts |
| OpenAI gpt-5-codex | ~330k mixed | ~55 prompts |
| Claude Haiku 4.5 | ~800k mixed | ~130 prompts |
| Gemini 2.5 Flash | ~1.5M mixed | ~250 prompts |
| Z.ai GLM-4.6 | ~1.8M mixed | ~300 prompts |
| MiniMax M2 / Qwen3-Coder | ~3.5M mixed | ~580 prompts |
| Gemini Flash-Lite | ~7M mixed | ~1,200 prompts |
| DeepSeek V4 Flash | ~8M mixed | ~1,400 prompts |
(Indicative only — real cost depends on your prompt size + cache hit rate.)
Token ≈ 4 characters in English; Burmese script ≈ 1 character per token (worse ratio — same text uses more tokens in Burmese).Input tokens= your message + the model's context (files, system prompt, history).** Output tokens**= what the model writes back.** Cache read**(Claude/OpenAI/Gemini) = repeated input cached at ~10% price. Huge saver in coding agents that re-send the same files.** A "long context"model (1M tokens) means you can paste a large codebase; you still pay per token. Tip:prefer agents that stream + truncate**old turns instead of re-sending everything.
Day 1, no card available:Gemini CLI(personal account) → free, instant. Day 1, with a card / Google account, want flexibility:****opencode + AI Studio API key→ free Flash-Lite tier, scale to paid later.** Premium quality, ok with spend:**Claude Code via Anthropic API or AWS Bedrock. Budget $5–10/student/mo, more for heavy use.Cheapest paid bulk:DeepSeek V4 Flash(~$0.14/$0.28 per M tokens) via opencode. Predictable flat:Z.ai GLM Coding Plan Lite($10/mo, works in Claude Code) or opencode Go($10/mo bundle).
The bootcamp will publish a shared proxy when it's deployed + secured — at that point students will get a single virtual key that routes to whatever model the proxy is configured for. Until then, pick from the list above.
Prices verified early-mid 2026. Confirm on each provider's official pricing page before quoting externally.Free-tier limits change. Google dropped 2.5 Pro free tier in April 2026; Flash dropped to ~250/day. OpenRouter's:free
rate-limit can change without notice.Cache assumptions matter. A 60% cache-hit rate is realistic for coding agents that revisit the same files; one-shot questions get ~0% cache and cost more.Quotas in prompts vs tokens are different things. Flat subs cap prompts; APIs cap tokens. A "120 prompts/5h" plan is not equivalent to "X tokens/5h" — measure your usage profile before committing.Burmese costs ~2× more in tokens than English for the same meaning. Plan for it on bilingual projects.
If you find a current price that disagrees with this table, the live provider page wins — open a PR to fix this doc.