Across 5 comparable benchmarks, Gemini 3.5 Flash-Lite has better reported results on 3 and worse on 2.
Moderate evidence5 of 5 shared results countedHow we compare
GPT-5.2-Codex is no longer available through its official API. API prices for this version are not shown.
Price basis: Standard rates vs listed rates. The price section below shows the applicable periods.
| Fact | Gemini 3.5 Flash-Lite | GPT-5.2-Codex |
|---|---|---|
| Context window | 1M tokens | — |
| Input / 1M | $0.300 | — |
| Output / 1M | $2.50 | — |
| Blended (4:1) | $0.546 | — |
| Status | Available | Retired |
| Released | July 21, 2026 | December 18, 2025 |
Blended is 4:1 input to output with 90% cache hits at listed rates — set your own mix.
Mean of 4 shared benchmarks grouped by capability on a 0–100 scale. How averages are built
Gemini 3.5 Flash-Lite ahead on 2 of 3 counted
Gemini 3.5 Flash-Lite ahead on 1 of 1 counted
No shared comparable results for Abstract reasoning, Mathematics, Science & health, Computer & tool use, Research & search, Long context, Professional work & productivity, Cybersecurity.
Published scores, with matching modes where available. Gaps show Gemini 3.5 Flash-Lite relative to GPT-5.2-Codex. Reading the results
| Benchmark | Gemini 3.5 Flash-Lite | GPT-5.2-Codex | Gap |
|---|---|---|---|
| APEX-SWERepository Engineering · Pass@1 | 31.2Thinking High · tools | 30.8Thinking High · tools | +0.4 pts#48 vs #51 of 61 |
| SWE-Bench Pro - PublicRepository Engineering · Accuracy | 54.2Thinking · tools | 41.0≠Unspecified Reasoning · tools | +13.2 pts#38 vs #60 of 66 |
| Vibe Code Bench v1.1Code Generation & Editing · Pass rate (%) | 37.2Thinking High · tools | 37.9Thinking High · tools | −0.8 pts#75 vs #73 of 121 |
| MMMU-ProVisual Understanding · Accuracy | 79.0Thinking · no tools | 76.3≠Extra-High · no tools | +2.7 pts#38 vs #58 of 136 |
| LiveBenchCross-capability Suites · Accuracy | 63.9Thinking High · no tools | 74.0≠Normal · no tools | −10.0 pts#73 vs #39 of 106 |
Repository Engineering · Pass@1
Repository Engineering · Accuracy
Code Generation & Editing · Pass rate (%)
Visual Understanding · Accuracy
Cross-capability Suites · Accuracy
A dash means no result is available.
| Benchmark | Gemini 3.5 Flash-Lite | GPT-5.2-Codex |
|---|---|---|
| CritPt | — | 8.7 |
| FrontierMath Tier 4 v2 | 0.0 | — |
| FrontierMath v2 | 26.0 | — |
| GPQA Diamond | 83.3 | — |
| Riemann-bench | 11.2 | — |
| SciCode | 41.3 | — |
| Code Migration | 6.2 | — |
| SWE-bench Multilingual | — | 66.3 |
| SWE-Bench Pro - Commercial | — | 27.7 |
| SWE-bench Verified | — | 72.8 |
| SWE-sweep | 0.12 | — |
| Terminal Bench Hard | — | 37.1 |
| Terminal-Bench 2.1 | 54.0 | — |
| Terminal-Bench 4.0 | 1.0 | — |
| APEX-Agents | 29.3 | — |
| OSWorld-Verified | 74.0 | — |
| Toolathlon-Verified | 57.1 | — |
| AA Intelligence Index v5.0.1 | 11.7 | — |
| ECI | 145 | — |
| Vals Index | 36.7 | — |
| Creative Writing | 1559 | — |
| IF Bench | — | 77.6 |
| Context Arena | 72.6 | — |
| GDP.pdf | 13.6 | — |
| AutomationBench | 4.0 | — |
| EMB (Excel Modeling) | 43.2 | — |
| Finance Agent v2 | 47.4 | — |
| Harvey's Legal Agent Benchmark | 0.0 | — |
| Legal Research Bench | 13.9 | — |
| Public Benefits Bench v1.1 | 51.8 | — |
| SAGE | 47.3 | — |
| Tax Agent Bench | 7.3 | — |
| τ³-Banking | 17.5 | — |
| ARC-AGI-1 | 53.5 | — |
| ARC-AGI-2 | 10.3 | — |
| MedCode | 43.5 | — |
| MedScribe | 70.9 | — |
Time-of-day providers use peak-hour rates; fixed-price providers use standard rates. Off-peak discounts remain in the published details.
Drag the handle above the bar for cache hits, the one below it for the input : output mix · 5% steps
Blended price = effective input rate at 90% cache hits × 80% + output rate × 20%
A blended cost needs a published input and output price on both sides, and one of these two does not have them yet.
Gemini 3.5 Flash-Lite: Standard pricing. Google Gemini API official pricing (ai.google.dev/gemini-api/docs/pricing) verified 2026-10-21; paid tier; standard tier
GPT-5.2-Codex: Listed text rates; pricing tier not confirmed.
| Price / 1M tokens | Gemini 3.5 Flash-Lite | GPT-5.2-Codex | Ratio |
|---|---|---|---|
| Input | $0.30 | — | — |
| Output | $2.50 | — | — |
| Cached input | $0.03 | — | — |
| Cache write | — | — | — |
| Blended | $0.546 | — | — |
Estimates use the selected token mix. Cache writes, tool fees and long-context surcharges are excluded.
Reported cost per task, paired with the score from the same run. Settings and sources may differ; costs do not affect the verdict.
| Benchmark | Gemini 3.5 Flash-Lite | GPT-5.2-Codex |
|---|---|---|
| AA Intelligence Index v5.0.1 | Not reported | |
| ARC-AGI-1 | Not reported | |
| ARC-AGI-2 | $0.139 / task Average · ARC-AGI-2 Score: 10.3 Thinking High (With Tools) $0.0437 / task Average · ARC-AGI-2 Score: 5.3 Thinking Medium (With Tools) | Not reported |
| AutomationBench | $0.13 / task Average · Pass Rate: 3.4 Thinking High (With Tools) · zapier-private-heldout $0.12 / task Average · Pass Rate: 4.0 Thinking Medium (With Tools) · zapier-private-heldout | Not reported |
| SWE-sweep | Not reported | |
| Tax Agent Bench | $0.0815 / task Average · Accuracy (%, rubric × citation quality): 7.3 Thinking High (With Tools) | Not reported |
| Attribute | Gemini 3.5 Flash-Lite | GPT-5.2-Codex |
|---|---|---|
| Availability | Available | Retired |
| Provider | Google Deep Mind | OpenAI |
| Model type | Multimodal model | Coding model |
| First release | July 21, 2026 | December 18, 2025 |
| Context window | 1M tokens | — |
| Max output | 64K tokens | — |
| Architecture | Undisclosed | Undisclosed |
| Documented capabilities | Reasoning, Coding, Translation, Function calling, Tool use, Structured output, Web search, Long context | Reasoning, Coding |
| Input modalities | Text, Image, Audio, Video | Text, Image |
| Output modalities | Text | Text |
| Weights | Closed / API only | Closed / API only |
| Commercial use | Proprietary | Proprietary |
These are reported-score comparisons, not controlled head-to-head tests.
What people found comparing these models, or using either one. Comments also appear on the models' own pages.
Used Gemini 3.5 Flash-Lite or GPT-5.2-Codex? Sign in with Google to share what worked, what did not, and which API or setup you tested.
No community comments on either model yet.
Alternatives with the most benchmark overlap for this matchup.