NEWEST GEMINI FLASH1M CONTEXTMULTIMODAL + PDF65K MAX OUTPUTHALF PRICE UNTIL 2027
Gemini 3.8 Flash API Pricing
Gemini 3.8 Flash shipped on September 2, 2026, described by Google as its most intelligent Flash model, built for long-horizon software engineering, autonomous agents and complex enterprise workflows. It lands on the same rate card as 3.7 Flash and 3.6 Flash: $0.75/M input, $3.75/M output and $0.075/M cached input through December 31, 2026, then standard pricing of $1.50 / $7.50 / $0.15 from January 1, 2027. That is three consecutive Flash generations at one price, so upgrading costs nothing and the only question is capability.
Input - per 1M tokens
$0.75/M
$1.50 from Jan 1, 2027 -50%
Output - per 1M tokens
$3.75/M
$7.50 from Jan 1, 2027 -50%
Cached input - per 1M tokens
$0.075/M
$0.15 from Jan 1, 2027 -90%
Effective - agentic blend
$0.481/M
92/8 split - 82% cache
§ 01 / TERMINAL
Gemini 3.8 Flash API cost calculator.
Live calculator on Google's current Flash rate. Everything here doubles on January 1, 2027 — run the numbers twice if the workload outlives this year.
$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
§ 02 / SCENARIOS
Gemini 3.8 Flash cost on real workloads.
CODING AGENT
Codebase-scale migration
$0.061/task
LONG DOC
Reading 100-page contracts
$0.065/doc
RAG SUPPORT
Ticket triage
$0.00532/ticket
ASSISTANT
Research planning turn
$0.011/turn
Estimate · gemini-tokenizer-estimate · ≈3.85 chars/token Auto-counts as you type
This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.
Characters 841
Words 128
Tokens (estimated) 218 tokens
Cost as input · uncached $0.00016 USD
Cost as output · uncached $0.00082 USD
Cost as cached input $0.00002 USD
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| Gemini 3.8 Flash Current | $0.75 cache $0.075 | $3.75 | $0.481 agentic 92/8 | 1M | Google's newest Flash - agents and long-horizon coding |
| Gemini 3.7 Flash | $0.75 cache $0.075 | $3.75 | $0.481 deprecated · same · promo | 1M | Deprecated; its calls now route here |
| Gemini 3.6 Flash | $0.75 cache $0.075 | $3.75 | $0.481 same · promo | 1M | Same price, older capability |
| Gemini 3.1 Pro Preview | $2.00 cache $0.20 | $12.00 | $1.44 +3.0× | 1M | Google's frontier reasoning tier |
| Gemini 3.5 Flash-Lite | $0.30 cache $0.03 | $2.50 | $0.272 −43% | 1M | High-volume, simpler tasks |
| Gemini 3.1 Flash-Lite | $0.25 cache $0.025 | $1.50 | $0.18 deprecated · −2.7× | 1M | Cheaper Lite tier, retiring May 2027 |
| Claude Haiku 5.5 | $0.10 cache $0.01 | $0.50 | $0.0682 −7.0× | 1M | Anthropic's fast, high-volume tier |
| GPT-6.1 Terra | $1.00 cache $0.05 | $5.00 | $0.645 +34% | 1.05M | OpenAI's fast, efficient GPT-6 tier |
| DeepSeek V4.1 Flash | $0.30 cache $0.006 | $1.20 | $0.15 −3.2× | 1M | Budget reasoning and coding |
Gemini 3.8 Flash has published scores on 10 benchmarks we track, and leads 2 of them and shares first place on ARC-AGI-1. 4 of these scores come from an independent leaderboard; the rest are the lab's own reports.
| Benchmark | Score | Rank | Leader | Source |
|---|---|---|---|---|
| ARC-AGI-1 Reasoning | 98.5% ARC Prize verified, best result | #1 of 99 | Tied for first with 5 more | Official leaderboard ARC Prize Foundation |
| ARC-AGI-2 Reasoning | 89.2% ARC Prize verified, best result | #9 of 105 | GPT-6 Astra · 95% | Official leaderboard ARC Prize Foundation |
| GDP.pdf Vision and documents · Default setting | 35% all-pass rate | #1 of 7 | Gemini 3.8 Flash leads | Vendor-reported Google DeepMind |
| Terminal-Bench 2 Agents and computer use · Terminal-Bench 2.1 | 89.4% as reported | #3 of 17 | DeepSeek V4.1 Flash · 90.6% | Vendor-reported Google DeepMind |
| CharXiv Reasoning Vision and documents · No tools | 86.2% no tools | #2 of 9 | Kimi K3.1 Pro · 89% | Vendor-reported Google DeepMind |
| DeepSWE Coding | 73.7% as reported | #6 of 27 | Gemini 4 Argon · 77.9% | Vendor-reported Google DeepMind |
| ARC-AGI-3 Reasoning · Standard harness | 10.4% ARC Prize verified, best result | #6 of 24 | GPT-6 Astra · 62.7% | Official leaderboard ARC Prize Foundation |
| OSWorld 2 Agents and computer use · 2.0, partial score, batch tool on | 59% partial score, batch tool enabled | #1 of 2 | Gemini 3.8 Flash leads | Vendor-reported Google DeepMind |
| Terminal-Bench 4.0 Agents and computer use | 19.1% high effort, mini-SWE-agent harness, 330 trials, ±3.4 pts (95% CI) | #25 of 34 | Kimi K3.1 Pro · 65.7% | Official leaderboard Terminal-Bench (Stanford, Laude Institute) |
| Harvey Legal Agent Benchmark Professional work | 10% all-pass rate | #6 of 7 | Kimi K3.1 Pro · 23.7% | Vendor-reported Google DeepMind |
Gemini 3.8 Flash API pricing FAQ.
What the identical rate card across three Flash generations means, what changes on January 1, 2027, and what this model takes as input.
Q · 01 How much does Gemini 3.8 Flash cost? +
Google lists
$0.75/M input, $3.75/M output (thinking tokens included) and $0.075/M cached input through December 31, 2026. From January 1, 2027 those become $1.50, $7.50 and $0.15. Under this site's 92/8 agentic blend at an 82% cache-hit rate the effective rate today is $0.481/M. A free tier is available with lower limits.Q · 02 Is it more expensive than 3.7 or 3.6 Flash? +
No — the three are priced identically, to the cent, on the same dated schedule. Google has now put three consecutive Flash generations on one rate card, so an upgrade costs nothing and the decision is purely about capability. Google positions 3.8 Flash for long-horizon software engineering and autonomous agents, and calls 3.7 Flash its previous generation as of this release.
Q · 03 What changes on January 1, 2027? +
Every token rate doubles: input
$0.75 → $1.50, output $3.75 → $7.50, cached input $0.075 → $0.15. Cache storage also doubles, from $0.50 to $1.00 per million tokens per hour. This page stores the rate in effect today; the scheduled one is recorded alongside it and will be re-checked on the day rather than assumed — a vendor announcing a change is not the same as a vendor applying it.Q · 04 What can it take as input? +
Text, image, video, audio and PDF in; text out. The input limit is
1,048,576 tokens and the output limit 65,536. Caching, code execution, function calling, file search and grounding with Google Maps are supported; computer use is in preview. Batch, Flex and Priority inference are all available.Q · 05 Are thinking tokens billed separately? +
No. Google states the output price includes thinking tokens, so a reasoning-heavy call shows up as ordinary output on the bill. That matters when comparing against vendors that meter a separate reasoning rate.