The Vibe Father
Roundup
Let's be honest before we rank anything: there is no canonical PHP or Laravel benchmark with the authority SWE-bench Verified has for Python. SWE-bench is built from real Python repositories, so a top score there is a direct proxy for real Python skill. Nothing equivalent exists for PHP at the same level of trust. So this ranking leans on general coding ability, how well a model handles Laravel's idioms, and real-world reports — not one clean number. This site itself is a Laravel app, so we care about getting this right, and we still won't pretend a leaderboard settles it. The live board lives at /benchmarks (VCI = SWE 40 / TB 30 / LCB 30).
Read the following as informed guidance, then do the one thing that actually decides it: test the shortlist on your own Laravel app. Your package set, your conventions, your Laravel version matter more than any benchmark.
Why the Python benchmark still tells you something
General coding ability transfers. A model that fixes real bugs across real Python repos reasons well about control flow, multi-file structure, and not breaking the neighbor — competence that carries into PHP. So SWE-bench Verified isn't a PHP ranking, but it's a reasonable starting signal for which models reason well about real code. Treat it as a shortlist filter, not the final word. The numbers below are general, not PHP-specific — that caveat is the point.
| Model | SWE-bench Verified (general proxy) | Why it matters for PHP/Laravel |
|---|---|---|
 Claude Fable 5 | 95.0 | Top ceiling for hard cross-package refactors |
| Claude Opus 4.8 | 88.6 | The safe default — elite, half of Fable's price |
| Claude Sonnet 5 | 85.2 | High-volume controllers, models, Blade views |
 GPT-5.5 | 80.6 | Strong all-rounder, comfortable with artisan/Composer |
 Gemini 3.5 Flash | 79.3 | 167 tok/s — fast iteration on views and routes |
 DeepSeek V4 Pro | 77.6 | $0.435 / $0.87 per M — cheap high-volume CRUD |
 Qwen3.7 Max | 77.3 | 204 tok/s — fastest here, great for rapid edits |
| Gemini 3.1 Pro | 75.6 | Strong self-contained problem-solving (LCB 88.5) |
What actually matters for PHP and Laravel specifically
Two things swing Laravel work that a general board doesn't capture.
Framework idiom. Laravel is opinionated, and good Laravel isn't just working PHP — it's Eloquent relationships instead of raw queries, form requests instead of inline validation, service providers, jobs, events, and the "Laravel way" of doing things. A model can write correct PHP that a Laravel reviewer would reject on sight. The frontier models tend to know current Laravel conventions well; weaker ones drift toward generic PHP or, worse, patterns from three major versions ago. This is exactly what your own-repo test surfaces.
Blade and the N+1 trap. Blade templating has its own quirks — and its own footguns, like lazy-loading a relationship inside a loop and quietly generating an N+1 query storm. A model that reasons carefully about data access will eager-load; a careless one ships a page that works in dev and melts in production. Review generated Eloquent code for query behavior, not just correctness.
Speed versus ceiling for Laravel work
A lot of Laravel is fast, repetitive scaffolding — a resource controller here, a migration and model there, a Blade view to match. In that loop, tokens-per-second is a feature. Gemini 3.5 Flash at 167 tok/s and Qwen3.7 Max at 204 tok/s make routine CRUD pleasant, and their ceiling is plenty for it. Save the frontier models for the hard part: a tricky multi-guard auth flow, a queue architecture, a payments integration where getting it wrong costs real money.
Picks by Laravel workload
- Complex feature or cross-package refactor. Reasoning ceiling wins. Opus 4.8 as the default; Fable 5 for the genuinely hard changes that touch auth, billing, or the queue layer.
- High-volume CRUD and scaffolding. Sonnet 5, or DeepSeek V4 Pro for value — capable enough for controllers and models, cheap enough to run all day.
- Fast view and route iteration. Gemini 3.5 Flash or Qwen3.7 Max. Speed makes the tweak-refresh loop feel instant.
- Full-stack all-rounder. GPT-5.5 is a strong single choice — solid PHP and at home with artisan, Composer, and the shell.
How to actually test on your stack
This beats every roundup and takes an afternoon:
- Pick one representative task from your real backlog — a real feature, a real bug, a real refactor. Not a toy.
- Give the same decision-complete task to two or three candidates — same files in scope, same definition of done. See prompt engineering for coding agents for how to make the comparison fair.
- Judge on your criteria: is it idiomatic Laravel, does it avoid N+1s, does it pass your Pest/PHPUnit tests, and how did the speed feel?
- Repeat on a second, different task. One is a data point; two is a signal.
Bringing your own API keys makes this cheap — swap the model, rerun the task, compare, paying only for tokens. The Vibe Father lets you point different models at the same job for exactly this reason, but any setup that switches models runs the experiment.
Our honest bottom line for PHP and Laravel
Want one model and no thinking? Opus 4.8 — it writes clean, current Laravel and rarely surprises you. Grinding CRUD at volume? DeepSeek V4 Pro or Sonnet 5. Iterating fast on views? Gemini 3.5 Flash or Qwen3.7 Max. Hold it loosely — the packages you lean on and your Laravel version can flip the ranking. Trust the board as a shortlist, then let your own app cast the deciding vote. For the language with a cleaner answer see the best AI model for Python, for adjacent work see the best AI model for SQL and the best AI for API development, and the overall field is in the best coding model roundup. Live numbers at /benchmarks.
