Kohin Edwards

19, building PromptToProd 🛠️ We rank AI coding tools on pricing, free tiers & code ownership. Affiliate deals never change a score.

[Kohin Edwards @PromptToProdApp]

Our coding agent scores now split into 3 numbers:

Best coding results (/40): Codex, Cursor, Amp, Claude Code, all 36
Best cost (/15): OpenCode, 14
Best portability (/10): OpenCode, goose, Cline, Qwen Code, OpenHands

Claude Code ties for best on results. It loses on cost and lock-in.

Which column picks your daily driver?
https://t.co/lWhOR335IO

[Kohin Edwards @PromptToProdApp]

AI code review pricing is all over the place:

CodeRabbit: $24/user/mo, free on public repos
Greptile: $1/credit, 1 dev free
Cursor Bugbot: ~$1-1.50 per review
Claude Code Review: token-billed, Team+ only

Top score: CodeRabbit 82. What reviews your PRs?
https://t.co/zz7IWn3Txt

[Kohin Edwards @PromptToProdApp]

What a $20/mo subscription actually costs you (US card):

Stripe: ~$0.88
Stripe + Billing: ~$1.02
Paddle: $1.50
Stripe Managed Payments: ~$1.58
Lemon Squeezy: $1.60

The extra ~70¢ buys you sales tax and VAT handled. Worth it, or file it yourself?
https://t.co/AWH9ehmq9n

[Kohin Edwards @PromptToProdApp]

@stephenmruiz Rough list-price math at 1M MAU:

WorkOS: $0 (free tier covers 1M)
Supabase: ~$2,950
Firebase: ~$4,415
Clerk: ~$17,225

The gap grows from ~$1K to ~$17K. Fair ask, a 1M column is going in.
https://t.co/vXGwiLA4Fs

[Kohin Edwards @PromptToProdApp]

What auth costs at 100K monthly users (list prices, re-checked Oct 25):

WorkOS: $0
Better Auth / Auth.js: $0 per user
Supabase Auth: $25
Firebase: ~$275
Clerk: ~$1,025
Kinde: ~$1,534
Descope: ~$4,549

Auth0: sales quote

Which one are you on?
https://t.co/vXGwiLA4Fs

[Kohin Edwards @PromptToProdApp]

Best free AI coding agents by our score:

OpenCode 87
Codex 86 (comes with ChatGPT Free)
goose 84
Kilo Code 83
Cline 81

Gemini CLI isn't on the list. Its free sign-in stopped June 18 and Google now points free users to Antigravity CLI (75).

What's your $0 daily driver?
https://t.co/bZijSSNv4w

[Kohin Edwards @PromptToProdApp]

@a1254766624 Mostly no. When results tie, cost decides what you try first, but portability decides whether you stay. Cheap and easy to leave beats cheap and sticky.

[Kohin Edwards @PromptToProdApp]

Our coding agent scores now split into 3 numbers:

Best coding results (/40): Codex, Cursor, Amp, Claude Code, all 36
Best cost (/15): OpenCode, 14
Best portability (/10): OpenCode, goose, Cline, Qwen Code, OpenHands

Claude Code ties for best on results. It loses on cost and lock-in.

Which column picks your daily driver?
https://t.co/lWhOR335IO

[Kohin Edwards @PromptToProdApp]

@zyeutequebec Agreed, that's the real switching cost. Rules and tests in the repo travel with you; a vendor settings pane doesn't. Where your config lives should probably count toward the portability score.

[Kohin Edwards @PromptToProdApp]

Free tiers with a catch:

Vercel Hobby: personal, non-commercial only
Supabase Free: pauses after 1 week idle
Render Free: web services sleep after 15 min
Resend Free: 3,000 emails/mo, but 100/day
Railway Free: 30-day trial, then $1/mo of credit

Which one bit you first?
https://t.co/7pigJrZajR

[Kohin Edwards @PromptToProdApp]

@RohanBhanotAI Mostly model choice and license: open source, bring any model, and how easily you can walk away. Config files aren't scored separately yet, and they probably should be, since rules and setup are a big part of the real switching cost.

[Kohin Edwards @PromptToProdApp]

The $0 stack, best-scored free tier per layer:

Agent: OpenCode (87)
DB: Supabase (93), 50K MAU
Auth: Clerk (87), 50K retained
Email: Resend (88), 3K/mo
Analytics: PostHog (91), 1M events
Payments: Stripe (85), per-txn only

Which layer would you swap?
https://t.co/kopTvDZZYy

[Kohin Edwards @PromptToProdApp]

@arbuge Same feeling, the list grows weekly. goose is Block's open-source agent: free, bring your own model, which is why it scores high on cost and portability. https://t.co/nWWWRJHEml

[Kohin Edwards @PromptToProdApp]

If it helps narrow it down, our top 5: OpenCode 87, Codex 86, goose 84, Kilo Code 83, Cursor 81.

Claude Code is #12 at 79 but ties for best on coding results. Pick by the column you care about most: results, cost or portability.

[Alex Bugeja @arbuge]

Too many coding harnesses/environments man...

* Claude Code
* Codex
* Cursor
* VSCode (still going strong!)
* Opencode
* Pi
* Antigravity
* Plain old CLIs...

etc. etc.

spoilt for choice every time I start a new project.

[Kohin Edwards @PromptToProdApp]

@Sithira269 Cost per task is the honest number once you're on subs, since sticker price stops mattering. At 50B tokens a month, which harness are you running?

[Kohin Edwards @PromptToProdApp]

@thegmol Good call. SSO and SCIM are where per-MAU math falls apart for B2B, since a lot of providers bill per connection on top. That deserves its own column next time we re-check prices.

[Kohin Edwards @PromptToProdApp]

Bundling auth into the base plan is a clean way to dodge the per-MAU line. What does it cost once you pass the free tier?

[Kohin Edwards @PromptToProdApp]

Undo rate is a great column, and it's the one we can't get from pricing pages or docs. Would you count it per task or per session?

[Kohin Edwards @PromptToProdApp]

Unpopular ranking: Claude Code is #12 on our AI coding agents list (79/100).

#1 OpenCode 87
#2 Codex 86
#3 goose 84
#4 Kilo Code 83
#5 Cursor 81

What sinks it: no free tier + 4/10 on lock-in.

Too harsh or fair? (Day 9/90)
https://t.co/5D1ggPZk1U

[Kohin Edwards @PromptToProdApp]

@xingzhounotes Good call, that split is live now: every tool on the coding agents list shows coding results, cost and portability separately.
https://t.co/5D1ggPZk1U

[Kohin Edwards @PromptToProdApp]

Fair, and our new split backs you up: Claude Code ties for best on coding results (36/40, with Codex, Cursor and Amp). It drops on cost (no free tier) and portability (4/10). A pricing gripe, yes, but plenty of people pick on price.
https://t.co/5D1ggPZk1U

[Kohin Edwards @PromptToProdApp]

Worth noticing it was Claude Code and Codex. On our board they tie for best coding results (36/40 each).

The gap between them is price and lock-in, not capability.

[Alx - ✿ @indie_alx]

Epic on Unreal Engine 6: "Yes, we are vibe-coding things."

Claude Code and Codex rewrote a big chunk of the D3D12 rendering layer in about a week. Their own estimate was 3 to 6 weeks for one engineer.

Still "not real engineering"?

https://t.co/JpoZKkYlLq

[Kohin Edwards @PromptToProdApp]

@xingzhounotes Fair point. The 79 is a blend of subscores (the 4/10 lock-in is one), and a daily-driver user would weight cost and portability very differently than someone trying their first agent. Which would you weight heaviest for a daily work tool?