Benchmarks

AI-assisted development, finally reliable.

Cursor, Copilot, and Claude Code search your codebase like a basic word processor — they scan for matching text and miss most of the places that actually need to change. Momental gives your AI the full picture before anything is touched — in milliseconds.

+45%
AI quality score when building new features
↑ 9.0 / 10 vs 6.2 / 10 without Momental
65%
of affected files found when renaming a feature
+62pp vs just 3% with standard search
+53%
better architectural decisions when AI knows your team’s history
↑ 9.2 / 10 vs 6.0 / 10 without Momental

Benchmark 1 · Results

How many places does AI find when you change a feature?

When you change how a feature works — its name, its parameters, its output type — every place in the codebase that uses it needs updating too. Without Momental, AI tools find those places by scanning for matching text. With Momental, AI knows the full map of relationships. We ran 20 tasks. These are the results on rename tasks — the scenario where missed updates cause the most broken builds.

Files correctly identified as needing changes

Standard search
3%
Momental
65% ↑+62pp
Dimension
Standard search
Momental
Files found that need updating (rename tasks)
3%
65% ↑+62pp

Live Example · Real Codebase

Same question. Completely different answer.

Task: “Add a required parameter to a billing function — find every place in the codebase that calls it.” Run on the same production codebase as Benchmark 1. 1,130 source files.

Standard search · 3.8–10s · 37 raw lines · manual triage required
Momental · 422ms · 32 named callers · risk level + tests included
Without Momental Standard text search
Search: "recordAsync(" across all files
packages/api/src/services/gemini.service.ts:195
packages/api/src/services/gemini.service.ts:220
packages/api/src/services/gemini.service.ts:289
packages/api/src/services/agent12/adapters/claude.ts:797
packages/api/src/services/agent12/adapters/openai.ts:196
packages/api/src/services/agent12/adapters/openai.ts:222
… 30 more lines
37 lines · 3.8–10s
  • 1 line is the function definition itself — requires manual triage to remove
  • No grouping by caller — the billing service appears 9 times in separate lines with no context
  • Different implementations of the same function are mixed together — no way to tell which is which
  • No risk level, no test suggestions
With Momental Code relationship map
Look up “recordAsync” in code map
Critical 441 affected files downstream
api: GeminiService · 9 call sites · 8 methods
api: ClaudeAdapter.streamWithMessages · line 797
api: OpenAIAdapter · 3 call sites
api: EmbeddingService.embed · EmbeddingService.embedTextBatch
worker: separate implementation — 9 callers, independent impact
Run: embedding.test.ts · gemini.service.test.ts · agent12.test.ts
32 named callers · 422ms (9–24× faster)
  • Each result is a named function — zero manual triage
  • Risk level: critical — billing path, 441 files affected downstream
  • Exact tests to run: 3 specific test files identified automatically
  • A second implementation in the worker is flagged separately — won’t be missed
Standard search
Momental
Time to complete
3.8–10s (full codebase scan)
422ms · 9–24× faster
What you get back
37 lines · 1 is definition
32 callers, all named functions
Manual review required
Yes — 20 files, no grouping
None
Risk level
Unknown
Critical · 441 files affected downstream
Tests to run
Unknown
3 specific test files
Multiple versions of function
Mixed into flat list
2 separate implementations, each with its own impact analysis

Measured live · April 2026 · Momental’s production codebase · 1,130 source files. Standard search: full repository text scan, no path scoping. Momental: two API calls, 185ms + 237ms = 422ms total.

External Validation · 379k lines

Works on any codebase from day one.

We ran the same test on the Microsoft TypeScript compiler — 379,000 lines of code, cold start, never previously indexed. Same result.

Standard search · cold start · full codebase scan
Momental · ~524ms total
Metric
Standard search
Momental
Time to find all callers
3.8–10s
~524ms (~20× faster)
Risk level provided
No
Yes
Specific tests identified
No
Yes

20 of 20 tasks measured · April 2026 · 13m 16s total · Microsoft/TypeScript · 601 source files · 379k lines. Standard search: full repository, cold start, no path scoping. Momental: ~275ms lookup + ~227ms callers = ~502ms total.

Benchmark 2 · Results

Does AI make better decisions when it knows your team’s history?

When your team has already decided how to build something — which database pattern to use, how to handle billing, which internal tools to use — your AI should know that. Without Momental, it gives generic advice based on public documentation. With Momental, it recalls your team’s actual past decisions. We asked 5 real architectural questions. An independent AI graded each answer from 1 to 10.

Question asked
Without team history
With team history
“Add a database query — use raw SQL or the ORM?”
7.2 / 10
8.2 / 10 ↑+14%
“New AI call — run it inline or queue it as a background job?”
7.5 / 10
9.0 / 10 ↑+20%
“New AI feature — use the vendor API directly or the internal billing-aware wrapper?”
6.0 / 10 (misses your billing rules)
9.2 / 10 ↑+53%
“Add a new AI text-extraction feature — how should it be built?”
6.2 / 10 (misses your cost controls and rate limits)
9.0 / 10 ↑+45%

The biggest gain is on questions where the right answer depends on your team’s specific rules. The AI knows the general best practice — but not that your team decided to always use the internal billing-aware wrapper instead of calling the vendor API directly. Momental surfaces that past decision. With it, the AI scores 9.2/10. Without it, 6.0/10.

Methodology

How we score every task

Every task runs twice — once with standard search tools, once with Momental. We measure five things on every run. No cherry-picking. Same codebase, same questions.

Complete

Did the AI find every place in the codebase that needs to change? The correct answer is computed fresh each run, not hardcoded.

🛠

Works

Does the AI’s proposed change actually work? Verified by an independent AI judge since running a full compiler on every task would be too slow.

Speed

How long from receiving the task to a working answer. Momental adds a small lookup cost upfront but eliminates the back-and-forth caused by missed callers.

📈

Data volume

How much information the AI processes. Momental sends a focused summary of exactly the relevant code instead of dumping dozens of files.

Quality

An independent AI grades the answer 1–10: is it correct, does it follow good practices, did it miss anything, does it introduce any new bugs?

Fairness rules

  • Same AI model for all runs
  • Same task descriptions — only the context injected differs
  • Correct answer verified fresh each run, not hardcoded
  • All raw scores published with results
  • Codebase: the Momental monorepo (TypeScript, ~300k lines, real production code) + microsoft/TypeScript (379k lines, public OSS)

All numbers from a 20-task benchmark run completed April 2026 on one production codebase (Momental’s own monorepo). External validation on microsoft/TypeScript (379k lines, public OSS) completed April 2026 — see the TypeScript section above. Runner source code is available on request.

Under the hood

What Momental gives the AI

At the start of each session, Momental gives the AI five ways to understand your codebase — as a connected map of relationships, not just a list of files.

Find any function

Look up any function, class, or feature by name. Returns exactly where it’s defined across the whole codebase.

See all connections

A complete picture: everything that calls this function, everything it calls, and any linked past decisions.

Measure the impact

Before changing anything, understand how many files are affected, what the risk level is, and which tests to run.

Search by meaning

Find related code by concept, not just keyword. Finds relevant code even when the name is completely different.

Coordinate safely

Tell other AI agents which files you’re editing. Prevents conflicts when multiple agents work the same codebase at once.

What we measure

Beyond code — the full picture

Momental combines code understanding with organizational memory — your team’s past decisions, patterns, and choices. These benchmarks measure both.

🔍

Full codebase map

Every function, every caller, across the whole codebase — not just the file that’s currently open. Nothing gets missed because the AI didn’t search the right folder.

📈

Impact scoring

A risk rating before any code is changed. Know exactly which tests to run and which teams to notify — before anything breaks.

🧠

Team memory

Your team’s past decisions surface in context — not just what the code does, but why it was built that way and what the agreed rules are.

👥

Multi-agent coordination

Live file claiming and conflict detection. Multiple agents working the same repo without stepping on each other.

Your AI deserves better context.

Momental is live. Connect your codebase and see results in minutes.