AI Model Release Tracker

Gemini 3.8 Flash

Released

Gemini 3.8 Flash is an AI model released by Google on Wednesday, Sep 2 2026, 20 days after Gemini 3.7 Flash. It has a 1M token context window. Benchmark results (shown below) cover Humanity's Last Exam (Verified), BioMysteryBench, GDP.PDF, LVBench, CheatBench, CursorBench 4.0, and 12 more.

API pricing

InputWhat you pay for everything you send the model — your question, plus any documents or earlier conversation you include with it.
$0.75
Cached inputA reduced rate for text you send over and over. If every request starts with the same instructions or the same document, the provider keeps a copy ready and charges less to read it again.
$0.075
OutputWhat you pay for the text the model writes back. It is normally the dearer half: producing an answer costs more than reading one.
$3.75
  • Output prices include reasoning tokens.
  • Introductory pricing. From January 1, 2027, $1.50 per million input tokens, $7.50 per million output tokens and $0.15 for context caching apply.
USD per 1M tokensEvery price here is for one million tokens. A token is roughly three-quarters of a word, so a million tokens is about 750,000 words of text.Verified October 13, 2026Read from ai.google.dev. Click the check to open the source.

Available from

Google AI Studio—flex$0.375$1.8751M
Google AI Studio——$0.75$3.751M
Googleglobal—$0.75$3.751M
Googleus—$0.825$4.1251M
Google AI Studio—priority$1.35$6.751M
Googleglobalpriority$1.35$6.751M
USD per 1M tokensEvery price here is for one million tokens. A token is roughly three-quarters of a word, so a million tokens is about 750,000 words of text.Checked October 17, 2026Third-party rates from openrouter.ai, not the lab’s own list price.

Benchmarks

Coding

CursorBench 4.0Agentic coding — Cursor's own test of coding agents on ambiguous, multi-file tasks taken from real Cursor sessions — editing, refactoring, investigating a codebase, understanding what the user meant, managing jobs and following a design. Cursor runs each model at several reasoning efforts; each release here carries the score of its best listed effort. Scores aren't comparable with earlier CursorBench versions. Higher is better.via CursorBench
39.6%
#12 of 16
DeepSWE 1.1Agentic coding — Artificial Analysis' independent test of deep, agentic software-engineering work — the AI has to plan and carry out substantial coding tasks end to end. (Version 1.1 of the test.) Higher is better.
71%
#10 of 36
Next.js EvalsNext.js coding — Vercel's open eval of how well AI coding agents build and migrate real Next.js apps — measured as the share of tasks the agent completes successfully. Higher is better.
90%
#7 of 33

Terminal & CLI

Terminal-Bench 4.0Agentic terminal coding — Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? Version 4.0 recalibrated how much time, CPU and memory each task gets, removed eight tasks and fixed nineteen, so fewer runs fail for reasons that have nothing to do with the model. Scores are not comparable with earlier versions. Higher is better.
19.1%
#23 of 29
Terminal-Bench 2.1Agentic terminal coding — Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? Higher is better.
89.4%
#2 of 35

Agentic & tool use

OSWorld 2.0Agentic computer use — Can the AI actually operate a computer — clicking, typing, and using real apps — to finish tasks on its own? Version 2.0 is a harder, refreshed task set. Higher is better.
59%
#8 of 10

Reasoning & science

Humanity's Last Exam (Verified)Multidisciplinary reasoning — The re-checked edition of Humanity's Last Exam: the same extremely hard expert questions, minus the ones found to be flawed or wrongly answered. Scores on it run lower than on the original exam, so read the two as separate tests rather than a before-and-after. Higher is better.
54.9%
#1
BioMysteryBenchBiology — Real unsolved-style biology puzzles — the AI has to reason its way to an answer the way a research biologist would. The “hard” split contains the toughest cases. Higher is better.
56.5%hard
88.8%human solved
hard
#1
human solved
#2 of 3
ARC-AGI-2Abstract reasoning — Puzzle-style tests of abstract reasoning and pattern-finding — the kind of thing people find easy but AIs often struggle with. Higher is better.via BenchLM
89.2%
#8 of 37
LAB-Bench 2Biology — Everyday tasks from a working biology lab — reading protocols, interpreting figures and sequence data, and answering the practical questions a researcher hits at the bench. Higher is better.
86.2%
#2 of 3

Knowledge work

GDPval-AA v2Knowledge work — economically valuable knowledge work (v2, re-based Elo)
1545
#11 of 21

Finance

Finance Agent v2Agentic financial analysis — Tests the AI on real financial-analysis work, like digging through reports and making sound decisions. Higher is better.
61.4%
#2 of 12

Multimodal

GDP.PDFDocument comprehension — Real professional PDFs — filings, reports, technical documents — with questions an expert in that field would ask. Tests whether the AI reads the page as a document, layout and figures included, rather than as loose text. Higher is better.
35%
#1
LVBenchVideo understanding — Can the AI follow a very long video — up to an hour — and answer questions that need details from far apart in it? Higher is better.
87.1%
87.8%agentic
#2 of 4
agentic
CharXiv ReasoningChart reasoning — Can the AI read and reason about complex charts and figures, not just text? Higher is better.
86.2%
#4 of 16

Robustness

CheatBenchCheating rate — Measures how often AI agents try to cheat on difficult assignments, such as reading hidden answers, copying work or manipulating grading. The overall score gives equal weight to ten categories; the sycophancy category measures how far an agent shifts its beliefs toward a user's stated views. Scores describe each model in its tested agent setup, and task success is measured separately. Lower is better.via CheatBench
75.2%
#10 of 13

Community preference

threejsevalCommunity preference (Three.js) — Every model gets the same prompt — "the Eiffel Tower", "a glass fishbowl", "a robot arm picking toys into a box" — and builds a 3D scene in Three.js. Real people then see two scenes side by side, names hidden, and vote for the one they prefer. The votes become a chess-style Elo rating on threejseval.com, averaged across all the prompts. It measures whether the scene looks and moves right to a human eye, not whether the code passes a test. Higher is better.
1577
#10 of 27

Source: CursorBench, retrieved 28 September 2026 · Source: BenchLM, retrieved 28 September 2026 · Source: CheatBench, retrieved 25 September 2026. Other sources are identified on the linked benchmark pages.

Compare Gemini 3.8 Flash with

Gemini 3.8 Flash

Suggested comparisons

Gemini 3.8 FlashvsGemini 4 ArgonGemini 3.8 FlashvsGPT-6.1 SolGemini 3.8 FlashvsClaude Haiku 5.5Gemini 3.8 FlashvsMuse Spark OpenGemini 3.8 FlashvsGrok 4.7Gemini 3.8 FlashvsDeepSeek-V4.1-Flash-1017Gemini 3.8 FlashvsMistral Large 4Gemini 3.8 FlashvsKimi K3Gemini 3.8 FlashvsGLM-5.3-FlashGemini 3.8 FlashvsQwen3.8-Max-0902Gemini 3.8 FlashvsNemotron 3.5 Super VL

About

Gemini 3.8 Flash, released September 2, 2026, was the fourth upgrade to Google's workhorse Flash tier in four months and the one aimed squarely at coding, the area where the line had trailed Anthropic and OpenAI. At launch it scored 89.4% on Terminal-Bench 2.1 for agentic terminal work — the best figure in Google's launch comparison, ahead of Claude Opus 5 at 89.1% and GPT-5.6 Sol at 88.8% — and 71.0% on DeepSWE v1.1 for long-horizon software engineering, up from 65.3% for 3.7 Flash. The harder agentic sets moved with it without closing the gap: 19.1% on Terminal-Bench 4.0 against 51.8% for Opus 5, and 59.0% on OSWorld 2.0 against 75.4%. Google kept the introductory price its predecessor had launched at, $0.75 per million input tokens and $3.75 per million output, against a regular rate of $1.50/$7.50, and the 1M-token context window came across unchanged.

The clearer wins at release were outside code. It led Google's comparison table on financial analyst work (61.4% on Vals Finance Agent v2), on legal workflows (10.0% all-pass-rate on Harvey's Legal Agent Benchmark, against 6.7% for Opus 5), on chart reasoning without tools (86.2% on CharXiv Reasoning), on long video (87.8% on LVBench navigating agentically, 87.1% static), and across the science evaluations — 56.5% on the human-difficult split of BioMysteryBench, seven points clear of the next model, and 86.2% on LAB-Bench 2. Expert reasoning landed at 54.9% on the verified edition of Humanity's Last Exam, within half a point of the frontier models it was priced at a fraction of, and knowledge work at a GDPval-AA v2 Elo of 1545. Read together the release was a Flash model buying most of a frontier model's breadth at Flash prices, while conceding the long-horizon agentic tasks to the tier above it.

Frequently asked questions

Gemini 3.8 Flash was released by Google on Wednesday, Sep 2 2026.

All Google releases

32 tracked