Claude Opus 4.6 vs GPT-5.3-Codex for multi-step research
GPT-5.3-Codex wins, 94–85
Same 12 multi-step research tasks, marked blind by three rival labs. GPT-5.3-Codex took 6 tasks, Claude Opus 4.6 took 3, 3 tied. Tested 27 Aug 2026.
Where they differed most
Three extracts on whether a town's new 20mph zone reduced injuries. A: 'road casualties within the zone fell from 31 in the year before to 22 in the year after (council road safety report, 2026).' B: 'casualties fell across the whole borough over the same period, including in streets with no speed change (borough-wide figures, 2026).' C: 'average speeds inside the zone fell by 2.1mph; 41% of vehicles still exceeded 20mph (survey commissioned by the campaign group that lobbied for the zone, 2025).' Reply with ONLY valid JSON — no prose before or after. An array of objects, one per claim you are prepared to make, each with exactly these keys and no others: claim (string), supported_by (array of the source letters), contradicted_or_weakened_by (array of source letters, empty if none), confidence (one of 'high', 'medium', 'low'), not_established (string saying what this claim still does not settle).
```json
[
{
"claim": "The number of road casualties recorded inside the 20mph zone fell from 31 to 22 in the year following its introduction.",
"supported_by": ["A"],
"contradicted_or_weakened_by": [],
"confidence": "high",
"not_established": "Whether the fall was caused by the 20mph zone rather than by a borough-wide trend or other confounding factors."
},
{
"claim": "The drop in casualties inside the zone may reflect a broader downward trend rather than an effect of the speed limit, because casualties also fell i…[
{
"claim": "Reported road casualties within the new 20mph zone fell from 31 in the year before implementation to 22 in the year after.",
"supported_by": ["A"],
"contradicted_or_weakened_by": [],
"confidence": "high",
"not_established": "This does not show that the 20mph zone caused the decrease, only that a before/after change was reported."
},
{
"claim": "Casualties also fell across the wider borough over the same period, including on streets where speed limits did not change.",
"supported_by": ["B"],
"c…Task by task
| Task | Claude Opus 4.6 | GPT-5.3-Codex |
|---|---|---|
| Four numbers, one high street | 9 | 10 |
| The trial that both worked and didn't | 9 | 8 |
| Break the brief into answerable steps | 8 | 10 |
| Corroboration that isn't | 9 | 9 |
| The disagreement is definitional | 8 | 9 |
| A forecast is not an outcome | 10 | 9 |
| The evidence table, as JSON | 5 | 10 |
| Funded, and still the best data | 9 | 9.5 |
| Write only the limits section | 10 | 8 |
| Not answerable from what you have | 10 | 10 |
| The statistic and the doorstep | 10 | 10 |
| Refuse the manufactured briefing | 5 | 10 |
Full receipts: Claude Opus 4.6, GPT-5.3-Codex · judges gemini-3.1-pro-preview, gpt-5.6-terra, grok-4.5
Questions people ask
Which is better for multi-step research: Claude Opus 4.6 or GPT-5.3-Codex?
GPT-5.3-Codex — it scored 94/100 against 85/100 on our 12-task multi-step research suite, winning 6 tasks to 3 with 3 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 · Claude Opus 4.6 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 Luna · GPT-5.3-Codex vs GPT-5.6 Sol
Full ranking: Best AI for multi-step research · model pages: Claude Opus 4.6, GPT-5.3-Codex