Across 12 comparable benchmarks, Gemini 3.1 Flash-Lite has better reported results on 9 and worse on 2, with 1 tied.
Moderate evidence12 of 12 shared results countedHow we compare
Price basis: Standard rates. The price section below shows the applicable periods.
| Fact | GPT OSS 120B | Gemini 3.1 Flash-Lite |
|---|---|---|
| Context window | 128K tokens | 1M tokens |
| Input / 1M | $0.350 | $0.250 |
| Output / 1M | $0.750 | $1.50 |
| Blended (4:1) | $0.430 | $0.338 |
| Status | Available | Deprecated |
| Released | August 6, 2025 | May 7, 2026 |
Blended is 4:1 input to output with 90% cache hits at listed rates — set your own mix.
Mean of 5 shared benchmarks grouped by capability on a 0–100 scale. How averages are built
Gemini 3.1 Flash-Lite ahead on 1 of 1 counted
Gemini 3.1 Flash-Lite ahead on 2 of 3 counted
Gemini 3.1 Flash-Lite ahead on 1 of 1 counted
Gemini 3.1 Flash-Lite ahead on 1 of 1 counted
Split 1–1 of 2 counted
No shared comparable results for Abstract reasoning, Mathematics, Research & search, Long context, Cybersecurity.
Published scores, with matching modes where available. Gaps show GPT OSS 120B relative to Gemini 3.1 Flash-Lite. Reading the results
| Benchmark | GPT OSS 120B | Gemini 3.1 Flash-Lite | Gap |
|---|---|---|---|
| Terminal Bench HardAgentic Development · Accuracy | 23.5Thinking High · tools | 24.2≠Thinking · tools | −0.7 pts#88 vs #83 of 157 |
| CritPtScientific Reasoning · Score | 1.1Thinking High · no tools | 1.1≠Thinking · no tools | =#103 vs #103 of 134 |
| GPQA DiamondScientific Reasoning · Accuracy | 80.1Thinking · no tools | 81.8≠Thinking High · no tools | −1.7 pts#111 vs #100 of 200 |
| SciCodeScientific Computing · Score | 34.0Thinking High · no tools | 43.4≠Thinking · no tools | 0.78×#88 vs #69 of 98 |
| PinchBench v2Tool Orchestration · Average score (%) | 44.8Reported best | 80.5Reported best | −35.7 pts#46 vs #14 of 51 |
| GDP.pdfDocuments & Charts · Task pass rate (%) | 4.0Thinking High · no tools | 7.8≠Thinking · no tools | −3.8 pts#77 vs #67 of 96 |
| Harvey Lab-AALegal · Criterion pass rate (some entries report all-pass rate) | 13.9Thinking High · tools | 31.1≠Thinking · tools | −17.3 pts#39 vs #36 of 39 |
| τ³-BankingService Workflows · Score | 12.8Thinking High · tools | 9.7≠Thinking · tools | 1.32×#83 vs #94 of 120 |
| HLEKnowledge Exams · Accuracy | 19.0Thinking · tools | 8.6≠Reported best | +10.4 pts#107 vs #133 of 151 |
| AA Intelligence Index v5.0.1Capability Indices · Index (0-100) | 3.8Thinking High · tools | 6.2Thinking High · tools | 0.62×#62 vs #57 of 62 |
| ECICapability Indices · ECI score (capability index, higher is better) | 140Reported best | 144Reported best | 0.97×#97 vs #81 of 168 |
| LiveBenchCross-capability Suites · Accuracy | 46.1Normal · no tools | 61.7≠Thinking High · no tools | −15.6 pts#100 vs #79 of 106 |
Agentic Development · Accuracy
Scientific Reasoning · Score
Scientific Reasoning · Accuracy
Scientific Computing · Score
Tool Orchestration · Average score (%)
Documents & Charts · Task pass rate (%)
Legal · Criterion pass rate (some entries report all-pass rate)
Service Workflows · Score
A dash means no result is available.
| Benchmark | GPT OSS 120B | Gemini 3.1 Flash-Lite |
|---|---|---|
| AIME 2024 | 96.6 | — |
| AIME2025 | 97.9 | — |
| FrontierMath v2 | — | 27.7 |
| MathArena Apex | 1.0 | — |
| MMLU | 90.0 | — |
| MMLU-Pro | 79.0 | — |
| Aider-Polyglot | 41.8 | — |
| APEX-SWE | 3.3 | — |
| Code Migration | — | 4.6 |
| CodeForces | 2622 | — |
| LiveCodeBench | 87.8 | — |
| SWE-bench Verified | 60.1 | — |
| Terminal-Bench 4.0 | — | 0.5 |
| Vibe Code Bench v1.1 | — | 0.0 |
| APEX-Agents | 4.4 | — |
| MCP-Atlas | — | 57.1 |
| Pinch Bench | 60.6 | — |
| AA Intelligence Index (historical versions) | 24.1 | — |
| Vals Index | — | 15.5 |
| Creative Writing | 956 | — |
| IF Bench | 69.0 | — |
| MMMU-Pro | — | 75.5 |
| AA-AnalystAgent | — | 8.8 |
| AutomationBench | 0.15 | — |
| EMB (Excel Modeling) | — | 8.6 |
| Finance Agent v2 | — | 30.0 |
| Harvey's Legal Agent Benchmark | — | 0.0 |
| Legal Research Bench | — | 3.4 |
| SAGE | — | 49.5 |
| SimpleBench | 22.1 | — |
| Vending-Bench 2 | -21.5 | — |
| MedCode | — | 47.6 |
| MedScribe | — | 63.9 |
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%
GPT OSS 120B costs27%morethan Gemini 3.1 Flash-Lite per million tokens.Using standard rates at this token mix.
GPT OSS 120B: Standard pricing. Cerebras Shared Inference official pricing (gpt-oss-120b) verified 2026-11-13
Gemini 3.1 Flash-Lite: Standard pricing. Google Gemini API official pricing (ai.google.dev/gemini-api/docs/pricing) verified 2026-10-21; paid tier; standard tier
| Price / 1M tokens | GPT OSS 120B | Gemini 3.1 Flash-Lite | Ratio |
|---|---|---|---|
| Input | $0.35 | $0.25 | 1.40× |
| Output | $0.75 | $1.50 | 0.50× |
| Cached input | — | $0.025 | — |
| Cache write | — | — | — |
| Blended | $0.43 | $0.338 | 1.27× |
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.
GPT OSS 120B relative to Gemini 3.1 Flash-Lite
Summary: 1 of 1 shared benchmarks have matching recorded settings and sources. The average is a geometric mean of 1 positive-cost ratios, one equal weight per benchmark. This benchmark set can include related tests and indexes; it is not a workload cost estimate.
GPT OSS 120B 1.8× more expensive
GPT OSS 120B 1.8× more expensive
Effort, protocol, aggregation and source match; other run conditions may differ.
GPT OSS 120B
Gemini 3.1 Flash-Lite
| Benchmark | GPT OSS 120B | Gemini 3.1 Flash-Lite |
|---|---|---|
| AutomationBench | $0.01 / task Average · Pass Rate: 0.15 Thinking High (With Tools) · zapier-private-heldout | Not reported |
| Attribute | GPT OSS 120B | Gemini 3.1 Flash-Lite |
|---|---|---|
| Availability | Available | Deprecated |
| Provider | OpenAI | Google Deep Mind |
| Model type | Reasoning model | Multimodal model |
| First release | August 6, 2025 | May 7, 2026 |
| Context window | 128K tokens | 1M tokens |
| Max output | 128K tokens | 66K tokens |
| Total parameters | 117B | — |
| Active parameters | 5.1B | — |
| Architecture | Mixture of Experts (MoE) | Undisclosed |
| Documented capabilities | Reasoning, Multilingual | Reasoning, Multilingual, Tool use, Long context |
| Knowledge cutoff | — | 2025-01 |
| Input modalities | Text | Text, Image, Audio, Video |
| Output modalities | Text | Text |
| Weights | Open weights | Closed / API only |
| Licence | Apache 2.0 | — |
| Commercial use | Commercial use allowed | 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 GPT OSS 120B or Gemini 3.1 Flash-Lite? 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.