Best AI for coding / head-to-head
GPT-5.3-Codex vs GPT-5.6 Lunafor coding
GPT-5.6 Luna wins, 93–92
Same 18 coding tasks, marked blind by three rival labs. GPT-5.6 Luna took 3 tasks, GPT-5.3-Codex took 2, 13 tied. Tested 16 Aug 2026.
Where they differed most
Identify the race condition in this code, explain what interleaving causes it, and fix it. let cache = null; async function getConfig() { if (cache) return cache; const r = await fetch('/config'); cache = await r.json(); return cache; }
The race is a classic **check-then-act** around an `await`.
### What goes wrong
`getConfig()` does:
1. Check `if (cache)`.
2. If empty, start `fetch` and `await`.
Because `await` yields control, another caller can run in between.
### Problematic interleaving
Assume two callers: A and B.
1. A calls `getConfig()`, sees `cache === null`.
2. A starts `fetch('/config')`, hits `await` (pauses).
3. B calls `getConfig()`, still sees `cache === null`.
4. B also starts `fetch('/config')`, hits `await`.
Now you have **two in-flight requests**.
If…The race is caused by the `await` yielding control before `cache` is assigned. For example: 1. Call A checks `cache` — it is `null`. 2. Call A starts `fetch()` and pauses at `await`. 3. Call B runs before A finishes and also checks `cache` — still `null`. 4. Call B starts a second `fetch()`. 5. Both requests eventually assign to `cache`; the one that finishes last determines the final value. Thus, concurrent callers can cause duplicate requests and may even receive different results. Cache the in-flight promise instead: ```js let configPro…
Task by task
| Task | GPT-5.3-Codex | GPT-5.6 Luna |
|---|---|---|
| Duration parser | 9 | 9 |
| Find the bug | 10 | 10 |
| SQL without window functions | 10 | 10 |
| Infinite useEffect | 9 | 9 |
| Typed debounce | 10 | 10 |
| Leftmost binary search | 10 | 10 |
| Security review | 9 | 9 |
| Safe migration | 8 | 8 |
| Fix the code not the test | 10 | 10 |
| Race condition | 9 | 10 |
| Retry with backoff | 8 | 9 |
| Recursive type | 9 | 9 |
| Bash one-liner | 9 | 9 |
| Explain and cost | 10 | 9 |
| Regex with limits | 9 | 8 |
| Callback to async | 9 | 10 |
| Diagnose from a trace | 8 | 8 |
| Idempotency | 10 | 10 |
Full receipts: GPT-5.3-Codex, GPT-5.6 Luna · judges claude-sonnet-5, gemini-3.1-pro-preview, grok-4.5
Questions people ask
Which is better for coding: GPT-5.3-Codex or GPT-5.6 Luna?
GPT-5.6 Luna — it scored 93/100 against 92/100 on our 18-task coding suite, winning 3 tasks to 2 with 13 tied. Every answer was marked blind by three judges from three rival AI labs.
How was this tested?
Both models answered the identical published coding tasks. Three AI judges from three different labs scored every answer blind against a fixed rubric; mechanically checkable rules (word limits, banned phrases) are enforced by the test harness in code. The raw outputs and judge verdicts are downloadable.
More coding head-to-heads: GPT-5.6 Luna vs GPT-5.6 Sol · GPT-5.6 Luna vs GPT-5.6 Terra · GLM 5.2 vs GPT-5.6 Luna · GPT-5.5 vs GPT-5.6 Luna · GPT-5.6 Luna vs Qwen3.7 Max · Claude Opus 4.8 vs GPT-5.6 Luna
Full ranking: Best AI for coding · model pages: GPT-5.3-Codex, GPT-5.6 Luna