How long each system takes
Honest ranges for every system, for both paths and both plans. Each range comes from dated facts about real projects.
The short answer
Getting a game to run on your PC takes weeks to months. A full decomp takes months to years.
The spread is wide on purpose. Three things move the number most: the size of the game, the tools for its system, and what work already exists. A small game on a system with strong tools can run in a few weeks. A big game on a system with weak tools can take a year.
Every number on this page is a range. None of them is a promise. Read the limits at the end before you plan around them.
Running on PC
You stop when the game plays from start to end on a PC. Bugs may remain. This is the faster path on every system.
Full decomp
You stop when all the code is readable source that builds. This is what you need to port the game to other systems.
Every system on one chart
Each row is one system. The teal bar is a Max 20x plan (now called Ultra). The amber bar is a Max 5x plan (now called Max). A longer bar means less is known, not just more work.
The numbers
This table holds the same ranges as the charts. Click a system to see the dated facts behind its numbers.
| System | Path A: how hard | Path A, Max 20x | Path A, Max 5x | Path B: how hard | Path B, Max 20x | Path B, Max 5x | How sure |
|---|---|---|---|---|---|---|---|
| Nintendo 64 | 6 to 16 weeks | 3 to 9 months | 6 to 17 months | 1 to 3 years | medium | ||
| GameCube | 4 to 12 weeks | 2 to 7 months | 9 to 24 months | 2 to 5 years | medium | ||
| Wii | 3 to 10 weeks | 2 to 6 months | 5 to 18 months, of which 1 to 3 months is the port | 9 months to 3 years, port included | medium | ||
| Wii U | 3 to 10 weeks | 2 to 6 months | 6 to 18 months, after Path A | 1 to 3 years | medium | ||
| Nintendo Switch | 3 to 10 weeks | 2 to 6 months | 1 to 3 years, and a full byte match may never be reached | 2 to 8 years | medium | ||
| Nintendo Switch 2 | 3 to 10 weeks | 2 to 6 months | 6 months to 3 years, for behavior-verified code; nobody has a byte-matching setup yet | 1 to 6 years | low | ||
| NES | 1 to 4 weeks | 2 to 10 weeks | 6 to 20 weeks | 3 to 12 months | medium | ||
| Super Nintendo | 2 to 8 weeks | 5 to 20 weeks | 3 to 9 months | 7 to 22 months | medium | ||
| Game Boy | 1 to 5 weeks | 2 to 12 weeks | 4 to 9 months | 8 months to 2 years | medium | ||
| Game Boy Color | 1 to 6 weeks | 2 to 15 weeks | 5 to 12 months | 9 months to 2.5 years | medium | ||
| Game Boy Advance | 2 to 8 weeks | 5 to 20 weeks | 4 to 18 months | 9 months to 3 years | medium | ||
| Nintendo DS | 4 to 12 weeks | 10 to 30 weeks | 3 to 9 months | 7 months to 2 years | low | ||
| Nintendo 3DS | 2 to 8 weeks | 5 to 20 weeks | 1 to 3 years | 2 to 6 years | medium | ||
| Virtual Boy | 2 to 8 weeks | 5 to 20 weeks | 4 months to 1 year | 9 months to 2.5 years | low | ||
| PlayStation | 2 to 8 weeks | 5 to 20 weeks | 5 to 14 months | 10 months to 3 years | medium | ||
| PlayStation 2 | 4 to 12 weeks | 10 to 30 weeks | 12 to 40 weeks | 30 to 100 weeks | medium | ||
| PlayStation 3 | 4 to 12 weeks | 2.5 to 7 months | 1 to 3 years | 2.5 to 7 years | medium | ||
| PlayStation 4 | 2 to 8 weeks | 5 to 20 weeks | 1 to 3 years | 2.5 to 7 years | medium | ||
| PlayStation 5 | 2 to 10 weeks | 5 to 25 weeks | 1 to 3 years | 2.5 to 7 years | low | ||
| PSP | 3 to 10 weeks | 2 to 6 months | 6 to 18 months | 1 to 3 years | high | ||
| Xbox | 4 to 12 weeks | 2.5 to 7 months | 6 months to 2 years | 1 to 5 years | medium | ||
| Xbox 360 | 3 to 10 weeks | 2 to 6 months | 1.5 to 4 years | 3 to 10 years | medium | ||
| Dreamcast | 3 to 10 weeks | 7 to 25 weeks | 10 to 30 weeks | 25 to 80 weeks | medium | ||
| Saturn | 2 to 8 weeks | 5 to 20 weeks | 9 months to 3 years | 1.5 to 6 years | medium | ||
| Genesis | 1 to 6 weeks | 3 to 16 weeks | 6 to 20 weeks | 14 to 50 weeks | medium | ||
| 32X | 3 to 10 weeks | 7 to 25 weeks | 4 to 11 months | 9 months to 2.5 years | medium | ||
| Game Gear | 1 to 5 weeks | 3 to 14 weeks | 8 to 30 weeks | 5 to 17 months | medium | ||
| PC | 6 to 16 weeks | 3.5 to 9 months | 6 to 18 months | 14 months to 3.5 years | low | ||
| Amiga | 2 to 8 weeks | 5 to 20 weeks | 3 to 9 months | 7 months to 2 years | low | ||
| Arcade | 4 to 12 weeks | 10 to 30 weeks | 3 to 9 months | 7 to 22 months | medium | ||
| CD-i | 2 to 8 weeks | 5 to 20 weeks | 2 to 6 months | 4 to 18 months | medium |
How these numbers were made
Each guide reviewed ten or more real projects for its system. For every project we wrote down four dates: the first commit, the first release, the latest release and the latest commit. We also wrote down how many people worked on it and whether the developer says AI was used.
Then we looked for words from the developers. Some say how long a step took. Those statements count more than dates, because dates can mislead.
From these facts each guide sets three ranges for each path:
- Max 20x. One lead session and several workers running side by side.
- Max 5x. One lead session and one or two workers. Bulk work takes longer.
- One person, no AI. Taken from projects that started before agents were common.
Each range also gets a label: low, medium or high confidence. Most are low or medium. A range is only marked high when several dated projects agree.
Dated facts from published work
These come from write-ups by the people who did the work. They are the firmest numbers we found.
| Project | System | What happened | Source |
|---|---|---|---|
| Snowboard Kids | Nintendo 64 | Full matching decomp in 84 days. Agents were used from day one. About 1 in 20 matches still needed an expert. | Chris Lewis |
| Snowboard Kids 2 | Nintendo 64 | 596 days in all. About a quarter was done by hand in the first 14 months. Agents did the rest in about 6 months. | Chris Lewis |
| Super Smash Bros. Melee | GameCube | After years of work it stood at 48.5% in January 2026. It reached 100% in September 2026, about 8 months later. | doldecomp/melee |
| Majora's Mask | Nintendo 64 | A team of people finished the decomp in about three years, from 2021 to December 2024. | 80 Level |
| A first PlayStation recomp | PlayStation | A beginner built a recompiler and got a first game playable in about three weeks with Claude Code. Later games took days to weeks. | 1379.tech |
| RecompOne | PlayStation | About five months of work, with no AI, before games could begin to be ported. | Read Only Memo |
| A PC shooter | PC | About three months with up to 16 agents. 83% of functions matched byte for byte. | momo5502 |
| Crimsonland | PC | A full rewrite from the game file in about two weeks. | banteg |
What these show
Agents cut the middle of the job the most. The first quarter is setup and the last tenth is the hard tail. Both still need you.
What makes it faster or slower
Size of the game
A game with 2,000 functions is a small job. A game with 20,000 is a large one. Count them in the Map phase before you guess a date.
Work that already exists
A finished decomp, a debug build with names, or a sister game on the same engine can save months. Check first.
Tools for the system
Some systems have years of tools behind them. Others have almost none. Each guide says which kind you have.
Your check
A strict check lets cheaper models do good work. A weak check means you redo work later. One team threw away a month this way.
Your plan
More workers finish the bulk phase sooner. They do not speed up the hard tail. That part waits on one strong session.
Your own hours
Someone must read reports, approve plans and review code. If you have one hour a day, the work moves at that pace.
What changes between the two plans
Max 20x (now called Ultra)
Run a lead session and several workers at once. The bulk phase moves fast. Watch your weekly limit: one author reports that an Opus loop left alone used up this plan in days.
Max 5x (now called Max)
Run a lead session and one or two workers. Expect the bulk phase to take two to four times longer. Another author reports hitting the limit often on this plan.
Anthropic does not publish exact usage numbers for either plan. Check /usage in Claude Code each day and adjust. See Claude Setup for the habits that stretch each plan.
Limits of this data
- Repo dates mislead. We checked all 1,699 projects that RecompLauncher lists. Most repos went public the same day as their first release. So "repo created" to "first release" says nothing about how long the work took. We used first commit dates and the developers' own words instead.
- Only winners are listed. Projects that stalled or failed are mostly not in the catalog. Real odds are worse than the list suggests.
- A first release is not a finished game. Many first releases boot and play, but still have bugs.
- No one publishes costs. We found no dollar figures for any agent-built project.
- Some systems have no published numbers. We found no first-hand write-up of an agent-built recomp for Xbox 360, PS2, PS3, GameCube, Wii or Switch. Those ranges come from repo dates and are marked low confidence.
- Models change. These facts are from 2024 to late 2026. Newer models may shift the ranges.
Sources
- Decompiling a Nintendo 64 game in 84 days
- The long tail of LLM-assisted decompilation
- Melee decomp, January 2026 progress
- Melee decomp, 100% in September 2026
- Majora's Mask decomp reaches 100%
- A PS1 static recompiler built with Claude Code
- Recomp and AI, five months later
- RecompOne interview
- Decompiling a PC game with agents
- Crimsonland rewrite
- RecompLauncher catalog