Gemini 3.8 Flash vs Step 5 Preview
Full profiles: Gemini 3.8 Flash (Google) and Step 5 Preview (StepFun): every benchmark, price and rank.
Updated · Data from 12 sources
As of 6 Nov 2026, Gemini 3.8 Flash is ahead of Step 5 Preview on quality (62.0 vs 54.9 on the BenchLeader Index), and they cost about the same at $1.50 vs $1.43 per 1M tokens blended. Choose Gemini 3.8 Flash for reasoning (64 vs 53), coding (60 vs 50), and knowledge (65 vs 55); choose Step 5 Preview for long context (69 vs 66). Gemini 3.8 Flash's current price runs until 31 Dec 2026, then $3.00.
Key facts (as of 6 Nov 2026)
- Gemini 3.8 Flash scores 62.0 on the BenchLeader Index and Step 5 Preview scores 54.9 (BenchLeader, 6 Nov 2026).
- Gemini 3.8 Flash costs $1.50 per 1M tokens blended until 31 Dec 2026, then $3.00 (https://ai.google.dev/gemini-api/docs/pricing, checked 20 Oct 2026).
- Step 5 Preview costs $1.43 per 1M tokens blended (list price, 6 Nov 2026).
- Of the 59 benchmarks both models have, Gemini 3.8 Flash leads 37 (BenchLeader, 6 Nov 2026).
Gemini 3.8 Flash benchmarks 
Step 5 Preview benchmarks Which to pick for…
The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.
| Use case | Pick | Why | Confidence |
|---|---|---|---|
| Coding | Gemini 3.8 Flash | 60 vs 50 category score; FrontierCode 2.0 (AA) 18.7% vs 12.0% | High |
| Agents & tool use | Either | 51 vs 51 category score; Arena Agent 2.6 vs -2.9 | Either |
| Maths | Gemini 3.8 Flash | 62 vs 60 category score; Arena Maths 1524 vs 1466 | Medium |
| Reasoning | Gemini 3.8 Flash | 64 vs 53 category score; HLE-Diamond (AA) 33.1% vs 24.3% | High |
| Knowledge | Gemini 3.8 Flash | 65 vs 55 category score; AA-Omniscience v1.1 23.3 vs 16.4 | High |
| Long documents | Step 5 Preview | 69 vs 66 category score; AA-LCR v1.1 88.3% vs 81.3% | Medium |
| Images and other media | Gemini 3.8 Flash | 67 vs 59 category score; Arena Vision 1315 vs 1265 | High |
| Chat quality (human preference) | Gemini 3.8 Flash | 69 vs 63 on Arena votes | Medium |
| Lowest cost | Either | $1.50 vs $1.43 per 1M blended; agentic coding session $0.060 vs $0.071; from 1 Jan 2027: $3.00 vs $1.43 | – |
| Fastest first answer | Either | 21.6 s (Gemini 3.8 Flash (high)) vs 25.9 s (Step 5 Preview) to the first answer token | – |
| High volume on a budget | Neither flagship | A smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs | – |
Key differences
- Gemini 3.8 Flash (high) leads on quality: 62.0 vs 54.9.
- Gemini 3.8 Flash (high) is stronger in agents & tools, coding, human preference, knowledge, maths, multimodal, reasoning.
- Step 5 Preview is stronger in long context, composite.
- They cost about the same ($1.43 per 1M blended) until 31 Dec 2026, then $1.43 vs $3.00 from 1 Jan 2027.
At a glance
| Spec | Gemini 3.8 Flash | Step 5 Preview |
|---|---|---|
| Developer | StepFun | |
| Released | 2 Sep 2026 | 18 Sep 2026 |
| Context window (tokens) | 1,048,576 | 1,000,000 |
| Compared at | Gemini 3.8 Flash (high) | Step 5 Preview |
| Open weights | No | Yes |
| Price in / out / cached input per 1M | $0.750 / $3.75 / $0.075 | $1.00 / $2.70 / $0.105 |
| Announced price change | $1.50 / $7.50 from 1 Jan 2027 | None announced |
Limits, availability and announced prices from the vendors' own pages: ai.google.dev pricing (checked 20 Oct 2026). Everything else is from the day's data.
Category by category
Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.
| Category | Gemini 3.8 Flash (high) | Step 5 Preview | Leader |
|---|---|---|---|
| Reasoning | 64.0 | 52.7 | Gemini 3.8 Flash |
| Coding | 60.3 | 50.2 | Gemini 3.8 Flash |
| Agents & tools | 51.1 | 50.6 | level |
| Maths | 62.1 | 59.7 | Gemini 3.8 Flash |
| Knowledge | 64.7 | 55.2 | Gemini 3.8 Flash |
| Human preference | 69.0 | 63.2 | Gemini 3.8 Flash |
| Multimodal | 67.4 | 59.4 | Gemini 3.8 Flash |
| Long context | 65.5 | 69.2 | Step 5 Preview |
| Composite | – | 40.8 | not comparable |
On one chart
Each domain on the index's own scale, where 50 is an average model. Click a domain to see the benchmarks behind it. Pick up to 5 models from the list.
Not drawn, measured for fewer than two of these models: Instruction following.
ModelIndex
Where sources disagree
Across the 59 benchmarks both have, Gemini 3.8 Flash leads 37, Step 5 Preview leads 20 and 2 are tied. Every overall index that rates both puts Gemini 3.8 Flash ahead.
| Overall index | Gemini 3.8 Flash (high) | Step 5 Preview | Leader |
|---|---|---|---|
| BenchLeader Index | 62.0 | 54.9 | Gemini 3.8 Flash |
| AA Intelligence Index v5.0 | 30.9 | 28.5 | Gemini 3.8 Flash |
| Vals Index | 56.0 | 54.8 | Gemini 3.8 Flash |
| Arena Text | 1496 | 1448 | Gemini 3.8 Flash |
Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).
Price, speed and latency
List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page. Cost, tokens and time per benchmark task are measured: Artificial Analysis’ averages over its whole Intelligence Index run.
| Measure | Gemini 3.8 Flash (high) | Step 5 Preview |
|---|---|---|
| Blended price per 1M tokens, until 31 Dec 2026 | $1.50 | $1.43 |
| Input / output per 1M | $0.750 / $3.75 | $1.00 / $2.70 |
| Blended price per 1M from 1 Jan 2027 | $3.00 | $1.43 |
| Input / output per 1M from 1 Jan 2027 | $1.50 / $7.50 | $1.00 / $2.70 |
| Cached input per 1M | $0.075 | $0.105 |
| Price per task: chat reply | $0.0014 | $0.0012 |
| Price per task: summarize a 30-page report | $0.011 | $0.014 |
| Price per task: code edit | $0.010 | $0.010 |
| Price per task: agentic coding session | $0.060 | $0.071 |
| Price per task: structured extraction | $0.0023 | $0.0025 |
| Agentic coding session, input mostly cached | $0.030 | $0.030 |
| Cost per benchmark task (AA Index) | $7.77 | $2.35 |
| Output tokens per task (AA Index) | 99k | 113k |
| Of which reasoning | 60k | 80k |
| Time per task (AA Index) | 14 min | 14 min |
| Output speed | 115 tok/s | 132 tok/s |
| Time to first answer | 21.6 s | 25.9 s |
| Time to a complete answer | 76.8 s | 57.3 s |
Per token is not per task: an average task of Artificial Analysis' Intelligence Index cost $7.77 on Gemini 3.8 Flash (high) and $2.35 on Step 5 Preview, and Step 5 Preview wrote 1.1× as many output tokens for the same work.
Each model is compared at its best-ranked reasoning-effort setting. Gemini 3.8 Flash (high) scores 62.0 and takes 22 s to its first answer token; Gemini 3.8 Flash (medium) scores 60.7 in 8 s, Gemini 3.8 Flash (low) scores 56.2 in 2 s. Step 5 Preview scores 54.9 and takes 26 s to its first answer token. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.
Every benchmark
148 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.
| Metric | Publisher | Gemini 3.8 Flash (high) | Step 5 Preview |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 62.0 | 54.9 |
| Agents & tools score | BenchLeader | 51.1 | 50.6 |
| Coding score | BenchLeader | 60.3 | 50.2 |
| Human preference score | BenchLeader | 69.0 | 63.2 |
| Knowledge score | BenchLeader | 64.7 | 55.2 |
| Long context score | BenchLeader | 65.5 | 69.2 |
| Maths score | BenchLeader | 62.1 | 59.7 |
| Multimodal score | BenchLeader | 67.4 | 59.4 |
| Reasoning score | BenchLeader | 64.0 | 52.7 |
| Composite score | BenchLeader | – | 40.8 |
| Blended price $/M | BenchLeader | $1.50 | $1.43 |
| Output speed | BenchLeader | 115 tok/s | 132 tok/s |
| Time to first answer | BenchLeader | 21.6 s | 25.9 s |
| Context window | BenchLeader | 1,048,576 tokens | 1,000,000 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 95.4% | – |
| Humanity's Last Exam | Scale AI / CAIS | 44.5%† | – |
| Arena Hard Prompts | Arena (LMArena) | 1517 | 1477 |
| LiveBench 2.0 Reasoning | LiveBench | 89.3%† | 83.8% |
| LiveBench 1.0 Reasoning | LiveBench | 89.3% | – |
| GPQA Diamond (AA) | Artificial Analysis | 95.3% | – |
| Humanity's Last Exam (AA) | Artificial Analysis | 47.8% | 46.5% |
| HLE-Diamond (AA) | Artificial Analysis | 33.1% | 24.3% |
| GPQA Diamond (Vals) | Vals AI | 94.4% | – |
| ARC-AGI-1 | ARC Prize | 98.5% | – |
| ARC-AGI-2 | ARC Prize | 89.2% | – |
| ARC-AGI-3 | ARC Prize | 35.0% | – |
| CritPt | Artificial Analysis | 18.3% | 20.9% |
| CritPt v2 hard subset (AA) | Artificial Analysis | 46.5% | 32.5% |
| MysteryMechanism | Vals AI | 36.5% | 18.5% |
| Chess Puzzles | Epoch AI Benchmarking Hub | 58.0% | – |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | 47.0% | – |
| LMCA | Epoch AI Benchmarking Hub | 51.5%† | – |
| DTBench | Epoch AI Benchmarking Hub | 95.7%† | – |
| Coding | |||
| SciCode | SciCode | 56.6% | 58.9% |
| WeirdML | WeirdML | 84.8% | – |
| FrontierCode | Cognition | 41.2%† | – |
| Arena Coding | Arena (LMArena) | 1530 | 1506 |
| Arena WebDev | Arena (LMArena) | 1583 | 1571† |
| LiveBench 2.0 Coding | LiveBench | 82.2%† | 85.6% |
Show all 148 benchmarks
| Metric | Publisher | Gemini 3.8 Flash (high) | Step 5 Preview |
|---|---|---|---|
| Coding | |||
| LiveBench 1.0 Coding | LiveBench | 72.5% | – |
| LiveCodeBench | Vals AI | 89.5% | – |
| IOI | Vals AI | 56.9% | 38.1% |
| SWE-bench (Vals) | Vals AI | 80.0% | – |
| Code Migration | Vals AI | 36.5% | 39.0% |
| Code Migration v1.1 | Vals AI | 38.6% | 47.0% |
| ProgramBench | Vals AI | 1.0% | – |
| SWE-bench Multimodal | Vals AI | 40.8% | 40.8% |
| Vibe Code Bench 1-100 | Vals AI | 18.8% | – |
| Vibe Code Bench v1.1 | Vals AI | 78.7% | 70.4% |
| SWE Atlas: Codebase QnA | Scale AI SEAL | 47.0%† | – |
| SWE Atlas: Refactoring | Scale AI SEAL | 44.8%† | – |
| SWE Atlas: Test Writing | Scale AI SEAL | 53.7%† | – |
| Surface Evolver Bench | Epoch AI Benchmarking Hub | 76.9% | – |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | 73.8% | – |
| CursorBench | Cursor | 39.6% | – |
| ALE-Bench | Epoch AI Benchmarking Hub | 1269.7 | – |
| FrontierSWE | Epoch AI Benchmarking Hub | 19.6% | – |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 19.7% | 33.3% |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | 87.6% | – |
| Startrise Frontend | Startrise | 64.6%† | – |
| GameBench 2 Overall | DuelLab GameBench 2 | 59.0† | – |
| GameBench 2 by reasoning setting | DuelLab GameBench 2 | 64.2 | – |
| FrontierCode 1.1 (AA) | Artificial Analysis | 36.4% | 38.4% |
| FrontierCode 2.0 (AA) | Artificial Analysis | 18.7% | 12.0% |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | 19.1% | – |
| Remote Labor Index | Scale AI / CAIS | 5.8%† | – |
| APEX-Agents | Mercor | 64.3%† | – |
| Arena Agent | Arena (LMArena) | 2.6 | -2.9 |
| LiveBench Long Agentic | LiveBench | 16.5%† | 19.4% |
| LiveBench Browser | LiveBench | 45.1%† | 24.8% |
| LiveBench Agentic Coding | LiveBench | 54.2% | – |
| Terminal-Bench 2.1 (Vals) | Vals AI | 81.3% | – |
| HiL-Bench | Scale AI SEAL | 41.5%† | – |
| GDPval-AA v2.1, normalised score | Artificial Analysis | 46.8% | 54.3% |
| GDPval-AA v2.1 | Artificial Analysis | 1445 | 1594 |
| τ³-Banking (AA) | Artificial Analysis | 45.0% | – |
| ITBench SRE (AA) | Artificial Analysis | 52.5% | 55.6% |
| Analyst Agent (AA) | Artificial Analysis | 46.3% | 35.0% |
| APEX-Agents (AA) | Artificial Analysis | – | 38.0% |
| CUA-bench | Vals AI | 4.2% | – |
| CyberBench | Vals AI | 43.8% | – |
| Finance Agent v2 | Vals AI | 61.4% | 50.7% |
| Finance Agent v2.1 | Vals AI | 65.8% | 58.9% |
| Harvey's Legal Agent Benchmark v1.1 | Vals AI | 12.1% | 12.9% |
| Harvey's Legal Agent Benchmark v1 | Vals AI | 10.0% | 9.2% |
| Legal Research Bench | Vals AI | 38.9% | 40.4% |
| SkillsBench | Vals AI | 58.0% | – |
| SREBench | Vals AI | 11.1% | 0.4% |
| Tax Agent Bench | Vals AI | 32.1% | 27.3% |
| Terminal-Bench 4.0 (Vals) | Vals AI | 19.7% | 31.8% |
| Terminal-Bench Science | Vals AI | 8.6% | 2.9% |
| Time Horizon Index: KSP | Vals AI | 18.8% | – |
| Vending-Bench 2 | Andon Labs | 5093.8† | – |
| AutomationBench | Artificial Analysis | 59.9% | 51.0% |
| GDP.pdf (AA) | Artificial Analysis | 21.0% | 14.8% |
| Harvey LAB-AA v1.1 | Harvey | 0.0% | – |
| EnterpriseOps-Gym | Artificial Analysis | – | 47.2% |
| AA-Briefcase v1.1 | Artificial Analysis | 1203 | 1425 |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | 10.0% | 2.4% |
| CWE-Bench (AA) | Artificial Analysis | 52.5% | 52.5% |
| DeepSec-Bench (AA) | Artificial Analysis | 14.5% | 15.8% |
| CyberGym end-to-end (AA) | Artificial Analysis | 8.4% | 74.8% |
| FACTS Search | 80.4%† | – | |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | 68.4% | – |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | 21.9% | – |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | 98.9% | – |
| ProofBench | Vals AI | 48.0%† | 42.0% |
| LiveBench Mathematics | LiveBench | 91.6% | – |
| Arena Maths | Arena (LMArena) | 1524 | 1466 |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 69.7% | – |
| LiveBench 2.0 Data Analysis | LiveBench | 54.0%† | 63.5% |
| LiveBench Research | LiveBench | 79.1%† | 72.8% |
| LiveBench 1.0 Data Analysis | LiveBench | 54.0% | – |
| AA-Omniscience v1.1 | Artificial Analysis | 23.3 | 16.4 |
| MMLU-Pro | Vals AI | 90.2% | – |
| LegalBench | Vals AI | 87.0% | – |
| TaxEval | Vals AI | 74.5% | – |
| PRBench Finance | Scale AI SEAL | 49.5%† | – |
| PRBench Legal | Scale AI SEAL | 48.9%† | – |
| BioMysteryBench | Vals AI | 62.2% | 67.4% |
| Excel Modeling Benchmark | Vals AI | 72.2% | 57.1% |
| Excel Modeling Benchmark v1.1 | Vals AI | 72.9% | 69.5% |
| MedCode | Vals AI | 48.1% | – |
| MedScribe | Vals AI | 84.5% | – |
| Public Benefits Bench v1.1 | Vals AI | 65.3% | 58.1% |
| GDP.pdf | Epoch AI Benchmarking Hub | 23.2% | – |
| AA-Omniscience v1.1: accuracy | Artificial Analysis | 50.5% | 36.4% |
| AA-Omniscience v1.1: non-hallucination | Artificial Analysis | 44.9% | 68.7% |
| HealthBench Professional (AA) | OpenAI | 51.7% | – |
| Instruction following | |||
| LiveBench Language | LiveBench | 87.8% | – |
| LiveBench Instruction Following | LiveBench | 81.4% | – |
| Arena Instruction Following | Arena (LMArena) | 1491 | 1446 |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1496 | 1448 |
| Arena Creative Writing | Arena (LMArena) | 1487 | 1411 |
| Arena Multi-turn | Arena (LMArena) | 1505 | 1442 |
| Arena Longer Queries | Arena (LMArena) | 1512 | 1460 |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | 1315 | 1265 |
| LiveBench Design | LiveBench | 72.7%† | 76.9% |
| MMMU-Pro | Artificial Analysis | 85.6% | 76.4% |
| MMMU-Pro (Vals) | Vals AI | 89.1% | – |
| MortgageTax | Vals AI | 65.3% | – |
| SAGE | Vals AI | 88.2% | – |
| AudioMC | Scale AI SEAL | 60.4% | – |
| AudioMC (text output) | Scale AI SEAL | 60.4% | – |
| Blueprint-Bench 2 | Epoch AI Benchmarking Hub | 38.6%† | – |
| Code2Video-AA | Artificial Analysis | 899 | 821 |
| AA-3D | Artificial Analysis | 966 | 854 |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 81.3% | 88.3% |
| MLCR | Artificial Analysis | 10.6% | 18.3% |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 158.6† | – |
| LiveBench 2.0 | LiveBench | 59.2%† | 58.6% |
| LiveBench 1.0 | LiveBench | 75.8% | – |
| AA Intelligence Index v5.0 | Artificial Analysis | 30.9 | 28.5 |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 40.9 | 43.7 |
| AA Progress Index v1 | Artificial Analysis | 320.3 | 318.2 |
| Vals Index | Vals AI | 56.0 | 54.8 |
| Vals Finance Index | Vals AI | 60.5% | 55.4% |
| Vals RSI Index | Vals AI | 19.9% | – |
| Vals Science Index | Vals AI | 35.8% | 29.6% |
| Vals Router Index | Vals AI | 56.0%† | – |
| AA Cyber Index | Artificial Analysis | 25.1 | 47.7 |
| Metadata | |||
| AA Openness Index | Artificial Analysis | – | 33.3 |
Data as of 2026-11-06. Best configuration of each model; every score links to its source on the model pages.
DuelLab GameBench 2, release gb2-20261103-09b2218cbd85 (https://benchmarks.duellab.org/releases/gb2-20261103-09b2218cbd85/), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Scores mapped to BenchLeader model names; not endorsed by DuelLab.
What changed
- Gemini 3.8 Flash: list price moves to $1.50 / $7.50 per 1M input / output tokens (announced).
- Gemini 3.8 Flash (high): 2 score revisions.
- Gemini 3.8 Flash (high): 1 new result (MirrorCode).
- Step 5 Preview: 2 new results (Vals Index, Harvey's Legal Agent Benchmark v1.1).
- Gemini 3.8 Flash (high): 1 score revision.
- Step 5 Preview: 1 score revision.
- Gemini 3.8 Flash (high): BenchLeader Index 63.3 → 61.8.
- Step 5 Preview: BenchLeader Index 58.0 → 54.7.
- Gemini 3.8 Flash (high): 5 new results (AA Intelligence Index v5.0, AA Progress Index v1, HLE-Diamond (AA)…); 3 score revisions; cost per task moved from $3.49/M to $7.77/M (123% up).
- Step 5 Preview: 5 new results (AA Intelligence Index v5.0, AA Progress Index v1, HLE-Diamond (AA)…); 2 score revisions; Step 5 Preview cost per task moved from $1.03/M to $2.35/M (128% up).
Related comparisons
- Gemini 3.8 Flash vs Claude Opus 5.5
- Gemini 3.8 Flash vs Gemini 4 Carbon
- Gemini 3.8 Flash vs GPT-6 Astra
- Step 5 Preview vs Gemini 4 Argon
- Step 5 Preview vs Muse Spark 1.3
- Step 5 Preview vs DeepSeek V4.1 Flash 1017
- Gemini 3.8 Flash vs Claude Fable 5.1
- Gemini 3.8 Flash vs Claude Fable 5
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
Gemini 3.8 Flash vs Step 5 Preview: questions
- Is Gemini 3.8 Flash better than Step 5 Preview?
- Gemini 3.8 Flash (high) leads on quality: 62.0 vs 54.9. The BenchLeader Index combines every independent quality benchmark; Gemini 3.8 Flash (high) is ahead overall as of 6 Nov 2026, but check the category scores for your use.
- Gemini 3.8 Flash vs Step 5 Preview: which is cheaper, and will that change?
- Step 5 Preview is cheaper today: $1.43 against $1.50 per million tokens, blended at three input tokens per output token. Gemini 3.8 Flash's current price runs until 31 Dec 2026, then $3.00.
- Is Gemini 3.8 Flash or Step 5 Preview better for coding?
- Gemini 3.8 Flash scores higher in coding (60 vs 50 on the category index, where 50 is average). On FrontierCode 2.0 (AA), the top shared coding benchmark, it is 18.7% for Gemini 3.8 Flash against 12.0% for Step 5 Preview.
- Which is faster, Gemini 3.8 Flash or Step 5 Preview?
- Step 5 Preview streams faster (132 vs 115 output tokens per second, Artificial Analysis 7-day median).
- Is Gemini 3.8 Flash or Step 5 Preview better for agents and tool use?
- They are level in agentic tasks (51 vs 51 on the category index). On Arena Agent, the top shared agents & tools benchmark, it is 2.6 for Gemini 3.8 Flash against -2.9 for Step 5 Preview.
More questions (4)
- Which is better for long documents?
- Step 5 Preview scores higher on long-context benchmarks (69 vs 66). Both take about 1M tokens: Gemini 3.8 Flash 1,048,576, Step 5 Preview 1,000,000.
- Which has lower latency?
- Gemini 3.8 Flash answers first: 21.6 s against 25.9 s to the first answer token, as Artificial Analysis measures it. Reasoning effort drives most of this: a higher setting thinks longer before it answers.
- Do independent benchmarks agree on Gemini 3.8 Flash vs Step 5 Preview?
- Across the 59 benchmarks both have, Gemini 3.8 Flash leads 37, Step 5 Preview leads 20 and 2 are tied. Every overall index that rates both puts Gemini 3.8 Flash ahead.
- When was Step 5 Preview released?
- Step 5 Preview was released on 18 Sep 2026.