GPT-5.3-Codex vs GPT-5.6 Luna for multi-step research
GPT-5.3-Codex wins, 94–90
Same 12 multi-step research tasks, marked blind by three rival labs. GPT-5.3-Codex took 4 tasks, GPT-5.6 Luna took 4, 4 tied. Tested 27 Aug 2026.
Where they differed most
Three sources on GP access in one integrated care board area. Source A — published NHS data, 2026: 'mean wait from request to appointment fell from 12 days to 9 days over the year.' Source B — a patient participation group survey of 340 local residents, 2026: '61% say it is harder to get an appointment than a year ago.' Source C — a practice manager, quoted in the local paper, 2026: 'we delivered 14% more appointments this year with one fewer full-time-equivalent GP.' Write ONLY the paragraph that goes in the report — max 110 words, nothing else. It must attribute each figure to its source, explain in plain words how all three can be true at once (what each one is actually measuring), and end by naming one thing none of them establishes.
Published NHS data for 2026 (Source A) show the mean wait from request to appointment fell from 12 to 9 days, measuring average elapsed time for appointments that were actually delivered. The 2026 patient participation group survey of 340 residents (Source B) found 61% felt access was harder than a year earlier, measuring people’s experience (including failed attempts, inconvenient slots, or not seeing a preferred clinician). The practice manager’s local-paper quote (Source C) that 14% more appointments were delivered with one fewer FTE GP meas…
NHS data (Source A) report that the average wait from requesting an appointment to receiving one fell from 12 days to nine. However, a patient participation group survey of 340 residents (Source B) found that 61% felt appointments were harder to obtain than a year earlier: perceptions may reflect difficulty reaching the practice, securing a preferred clinician or getting an appointment when needed, not the average wait. Meanwhile, a practice manager told the local paper (Source C) that the practice delivered 14% more appointments despite having…
Task by task
| Task | GPT-5.3-Codex | GPT-5.6 Luna |
|---|---|---|
| Four numbers, one high street | 10 | 9 |
| The trial that both worked and didn't | 8 | 9 |
| Break the brief into answerable steps | 10 | 10 |
| Corroboration that isn't | 9 | 9 |
| The disagreement is definitional | 9 | 9 |
| A forecast is not an outcome | 9 | 10 |
| The evidence table, as JSON | 10 | 8 |
| Funded, and still the best data | 9.5 | 10 |
| Write only the limits section | 8 | 10 |
| Not answerable from what you have | 10 | 10 |
| The statistic and the doorstep | 10 | 5 |
| Refuse the manufactured briefing | 10 | 9 |
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 multi-step research: GPT-5.3-Codex or GPT-5.6 Luna?
GPT-5.3-Codex — it scored 94/100 against 90/100 on our 12-task multi-step research suite, winning 4 tasks to 4 with 4 tied. Every answer was marked blind by three judges from three rival AI labs.
How was this tested?
Both models answered the identical published multi-step research 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 multi-step research head-to-heads: GPT-5.3-Codex vs GPT-5.6 Terra · GPT-5.6 Luna vs GPT-5.6 Terra · Claude Sonnet 5 vs GPT-5.3-Codex · GPT-5.3-Codex vs GPT-5.5 · GPT-5.3-Codex vs GPT-5.6 Sol · GPT-5.3-Codex vs Kimi K3
Full ranking: Best AI for multi-step research · model pages: GPT-5.3-Codex, GPT-5.6 Luna