← All models

Gemini 3.8 Flash

Language Available Full comparison ↗

google-deepmind/gemini-3.8-flash · by Google DeepMind · decoder-only-transformer

Pricing — 1 offering(s)

Google DeepMind

Input tokens

  • $0.75 / 1M tokens (input) Current 2026-09-02 → 2026-12-31
    Introductory rate through Dec 31 2026; scheduled step-up to $1.50/1M on Jan 1 2027 per Google's published…

    Introductory rate through Dec 31 2026; scheduled step-up to $1.50/1M on Jan 1 2027 per Google's published pricing page — not yet in effect, no forward-dated entry added ahead of it.

  • $1.50 / 1M tokens (input) Future 2027-01-01

Cached input tokens

  • $0.075 / 1M tokens (cached input) Current 2026-09-02 → 2026-12-31
    Introductory rate, $0.075/1M through 2026-12-31; $0.15/1M from 2027-01-01. Storage is billed separately…

    Introductory rate, $0.075/1M through 2026-12-31; $0.15/1M from 2027-01-01. Storage is billed separately ($0.50/1M tokens per hour through 2026, $1.00 from 2027).

  • $0.15 / 1M tokens (cached input) Future 2027-01-01

Output tokens

  • $3.75 / 1M tokens (output) Current 2026-09-02 → 2026-12-31
    Includes thinking tokens. Introductory rate through Dec 31 2026; scheduled step-up to $7.50/1M on Jan 1 2027…

    Includes thinking tokens. Introductory rate through Dec 31 2026; scheduled step-up to $7.50/1M on Jan 1 2027 per Google's published pricing page — not yet in effect, no forward-dated entry added ahead of it.

  • $7.50 / 1M tokens (output) Future 2027-01-01

Batch

Batch input tokens

  • $0.375 / 1M tokens (input) Current 2026-11-03 → 2026-12-31
  • $0.75 / 1M tokens (input) Future 2027-01-01

Batch output tokens

  • $1.875 / 1M tokens (output) Current 2026-11-03 → 2026-12-31
  • $3.75 / 1M tokens (output) Future 2027-01-01

Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.

See how Gemini 3.8 Flash fits into a cost-aware routing setup

See how →

Capability profile

How Gemini 3.8 Flash rates across core capability dimensions, with the task-level evidence behind each rating.

reasoning Strong

HLE-Verified (expert-level multi-step reasoning across STEM, humanities, and professional fields): 54.9%. Vendor-reported (Google DeepMind model card) — narrowly ahead of Claude Opus 5's reported 54.4% on the same benchmark, notable for a Flash-tier model.

coding Strong

Terminal-Bench 2.1 (agentic coding tasks executed in a real terminal environment): 89.4% — fails roughly 1 task in 10. DeepSWE v1.1 (long-horizon software engineering): 73.7%. Both vendor-reported (Google DeepMind model card, deepmind.google/models/model-cards/gemini-3-8-flash/). No SWE-bench Verified/Pro score published by Google for this release. Independent (Vals AI): SWE-bench Verified 80.0% (mini-swe-agent), Terminal-Bench 2.1 81.3% (Terminus 2, below Google's 89.4%), Terminal-Bench 4.0 19.7% and LiveCodeBench 89.5%.

tool use Strong

Gemini API docs (ai.google.dev/gemini-api/docs/models/gemini-3.8-flash): function calling, code execution, and file search documented as supported; computer use listed as supported (Preview).

instruction following Strong
context window Strong

1,048,576-token input context, 65,536-token output limit.

multilingual Strong
speed Strong
cost efficiency Strong

$0.75/$3.75 per 1M input/output tokens (introductory, through 2026-12-31) — cheaper on both ends than same-tier competitor GPT-5.6 Luna ($1.00/$6.00).

Coding benchmarks

View the full coding benchmark leaderboard →
Benchmark Score Harness Source
swe-bench-verified ↗ 80.0% 2026-09 mini-swe-agent 📊 leaderboard ↗
terminal-bench-2-1 81.3% 2026-09 terminus-2 📊 leaderboard ↗
livecodebench ↗ 89.5% 2026-09 — 📊 leaderboard ↗
terminal-bench-4-0 19.7% 2026-10 mini-swe-agent 📊 leaderboard ↗

Operator guidance

Google's current recommended default Flash-tier model as of 2026-09-02, superseding Gemini 3.5 Flash in this registry (Gemini 3.6/3.7 Flash were not separately tracked here). Step up to Gemini 3.1 Pro for the hardest reasoning/coding tasks; otherwise this is the cost/latency option in the current Gemini 3.x line.

Use cases

Limitations

Citations