Gemini 3.8 Flash
Google DeepMind · released
Google's current mainstream model: fast, low-cost, multimodal and built around a 1M-token context, aimed at coding agents and multi-step knowledge work. It ranks second on the Finance Agent benchmark despite sitting mid-table overall.
Best for: Multimodal input at low cost · Finance and document work · High-throughput agents
- Intelligence Index
- 40.9
- #17 of 38 · at High effort
- Price per 1M tokens
- $0.75 in $3.75 out
- Blended $1.5 at 3:1
- Context window
- 1M
- Up to 66K output
- Output speed
- 125 tokens/s
- $1.24 measured cost per task
Capability profile
One 0–100 score per area, placing this model between the weakest and strongest tracked model across that area's benchmarks.
- Overall70 · #10 of 38
Based on 3 of 3 benchmarks
- Coding51 · #19 of 39
Based on 5 of 5 benchmarks
- Agents & knowledge work71 · #11 of 40
Based on 6 of 6 benchmarks
- Reasoning & science66 · #12 of 39
Based on 2 of 2 benchmarks
- Knowledge & reliability83 · #8 of 39
Based on 2 of 2 benchmarks
- Vision92 · #4 of 31
Based on 2 of 2 benchmarks
Reasoning effort trade-off
The same model at each effort setting. Lower settings give up a few points for a fraction of the cost and wait.
| Setting | Intelligence Index | Cost / task | First answer |
|---|---|---|---|
| High | 40.9 | $1.24 | 24s |
| Medium | 39.8 | $0.93 | — |
| Low | 33.5 | — | — |
Measured by Artificial Analysis. First answer is the median wait until the first answer token, including thinking time.
Benchmark results
All third-party results on record. Rank is among the tracked models that have been tested on that benchmark.
Overall
#17 of 38· High effort· Artificial Analysis
- Vals Index54.8%
#10 of 26· High effort· $5.73 per test· Vals AI
- Arena · Text1495 ±5
#4 of 33· High effort· 26,298 votes· Arena
Coding
- Terminal-Bench 4.019.7%
#20 of 38· High effort· Artificial Analysis
- SciCode56.6%
#12 of 38· High effort· Artificial Analysis
- Arena · WebDev1583 ±8
#19 of 34· High effort· 8,981 votes· Arena
- Vibe Code Bench78.7%
#16 of 26· High effort· Vals AI
- Code Migration31.6%
#19 of 26· High effort· Vals AI
Agents & knowledge work
- GDPval-AA46.8
#22 of 39· High effort· Artificial Analysis
- Arena · Agent+3.0 ±0.8
#13 of 29· High effort· Arena
- Finance Agent v261.4%
#2 of 26· High effort· Vals AI
#5 of 26· High effort· Vals AI
- Legal Research Bench38.9%
#18 of 26· High effort· Vals AI
- Tax Agent Bench66.8%
#8 of 26· High effort· Vals AI
Reasoning & science
- Humanity's Last Exam47.8%
#9 of 39· High effort· Artificial Analysis
- CritPt18.3%
#15 of 39· High effort· Artificial Analysis
- ARC-AGI-2saturating89.2%
#5 of 16· High effort· $0.40 per task· ARC Prize
- ARC-AGI-310.4%
#3 of 6· High effort· ARC Prize
- GPQA Diamondsaturating95.3%
#2 of 27· High effort· Artificial Analysis
Knowledge & reliability
#9 of 39· High effort· Artificial Analysis
- AA-LCR81.3%
#17 of 39· High effort· Artificial Analysis
Vision
- MMMU-Pro85.6%
#4 of 27· High effort· Artificial Analysis
- Arena · Vision1290 ±12
#3 of 19· High effort· 3,086 votes· Arena
Reported by Google DeepMind
Numbers the lab published at launch, run on its own harness and settings. Useful context, but not independently verified and never used in the rankings on this site.
- DeepSWE v1.173.7%
- OSWorld 2.0 · partial score59%
- Terminal-Bench 4.0 · public leaderboard19.1%
Specifications
- Released
- Sep 2, 2026
- Availability
- Generally available
- Context window
- 1,048,576 tokens
- Max output
- 65,536 tokens
- Accepts
- Text, Images, Audio, Video, PDFs
- Knowledge cutoff
- Mar 2026
- Weights
- Closed
- Reasoning control
- Thinking level low to high
- API model ID
gemini-3.8-flash
Pricing
- Input / 1M tokens
- $0.75
- Cached input / 1M
- $0.07
- Output / 1M tokens
- $3.75
- Blended (3:1)
- $1.5
- Measured cost / task
- $1.24
Introductory price until 31 December 2026, then $1.50 in / $7.50 out.
Official sources
Closest alternatives
The models nearest to this one on overall score.
23 benchmarks tracked. See methodology and sources.