none reviewed yet · 2 benchmark observations · latest capture Sep 8, 2026
Inspect the evidenceGoogle · MODEL PROFILE
Gemini 3.8 Flash
Cost-sensitive teams that need agentic software engineering plus a free experimentation tier
Saved editorial fit for your current task lens; not a benchmark result
How scoring worksVerified rates (Sep 8, 2026) · 100 runs / month at your token settings
Adjust workloadAt a glance
9 facts- Provider
- Type
- API model
- Access
- Gemini API, Google AI Studio, Gemini Enterprise Agent Platform; Standard/Batch/Flex/Priority service tiers
- Context
- 1,048,576 tokens
- Maximum output
- n.a. on fetched Google pages
- License / weights
- Proprietary
- Input modalities
- text, image, video, file, audio
- Output modalities
- text
- Pricing notes
- Standard $0.75 in / $3.75 out per 1M through Dec 31, 2026, then $1.50/$7.50 from Jan 1, 2027; Batch and Flex $0.375/$1.875; Priority $1.35/$6.75; caching $0.075 plus $0.50 per 1M tokens/hour storage
The practical trade-offs
Documented strengths
- DeepSWE v1.1 73.8% is second-highest in OpenAI's own cross-vendor table, above Claude Fable 5.1 (67.4%)
- $0.75/$3.75 promotional pricing plus a real free tier
- Design Arena Website 1301 is the strongest of any sub-$1 input-price model except its own siblings
- Cyber sibling shows unusually strong vulnerability-patching evidence
Limitations to check
- Terminal-Bench 4.0 19.1% in OpenAI's table is far behind frontier agentic scores
- Arena frontend Elo (1569) sits below the older Gemini 3.7 Flash (1592)
- Google does not publish context window, max output or modalities on the model or pricing pages
- AA has no output-speed measurement yet
Follow the source
8 source linksEvidence & conditions
8Source links
NoneReviewed records
2Benchmark runs
Sep 8, 2026Latest capture
- Evidence source 1
ai.google.devDate not recorded - Evidence source 2
ai.google.devDate not recorded - Evidence source 3
ai.google.devDate not recorded - Evidence source 4
blog.googleDate not recorded - Evidence source 5
arena.aiDate not recorded - Evidence source 6
benchlm.aiDate not recorded - Evidence source 7
artificialanalysis.aiDate not recorded - Evidence source 8
openai.comDate not recorded
Benchmark observations
HLE-Verified Inspect conditions
- name
- HLE-Verified
- value
- 54.9
- unit
- %
- source Url
- https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/
- checked At
- 2026-09-08T19:30:56Z
- attribution
- provider-reported
- configuration
- Google report for standard Gemini 3.8 Flash, not Flash Cyber. Effort and tool budget not specified in extracted prose.
DeepSWE v1.1 Inspect conditions
- name
- DeepSWE v1.1
- value
- 73.8
- unit
- %
- source Url
- https://openai.com/index/gpt-6-astra/
- checked At
- 2026-09-08T19:30:56Z
- attribution
- provider-reported
- configuration
- OpenAI comparative table; maximum across effort; Google model harness not specified.
Source reports, measured results, and editorial assessments are separate evidence types. Source dates do not imply a new benchmark run. Capture dates are shown when recorded; otherwise we say the capture date was not recorded. Recorded dates are compared with this catalogue snapshot (Sep 11, 2026), not today’s date.
Method AIS-V2
How this page is assessed
Quality claims follow Method AIS-V2.
Money never buys rank — commercial relationships cannot change rank, eligibility, evidence quality, or publication timing.
Submit or claim a listing is free and does not skip the editorial queue.
Private · this device only