AI tools·Chat models·Global
Gemini
Google’s front door across Search, Gmail, Docs, YouTube, and huge context. The Flash line iterates very fast.
Compiled through 2026-09
- Vendor
- Started
- 2015
- China availability
- Overseas accessRequires overseas network access
- Pricing tier
- Free tier
- Current flagship
- Gemini 3.8 Flash / 2.5 Pro
- Website
- gemini.google.com
Evaluation, Pros & Cons
Editorial score based on public benchmarks and community feedback — not a third-party benchmark result
Key Advantages
- Up to 2M token native context window with near-zero needle-in-haystack loss
- Deep integration with Google Workspace, Docs, YouTube, and Android ecosystems
Limitations & Caveats
- Occasional tone variations in very long bilingual generations
Overview
Gemini is DeepMind’s multimodal flagship (successor to Bard), in the Gemini app, AI Studio, Vertex AI, Workspace, and Android. 1.5 made million-token context a headline; 2.x added agents. In 2026 the Flash line shipped rapidly (3.6, 3.7, then 3.8 Flash on Sep 2). Strong on docs, sheets, long video, and retrieval; coding still often trails Claude / GPT flagships.
Direction
The public strategy is models inside every Google surface: Search, Chrome, Android, IDEs such as Antigravity, and Vertex. Flash is cheap and frequent; Pro is deeper reasoning. Multimodal and long context stay core; coding agents are the catch-up area.
Version history
Sep 2026
LatestGemini 3.8 Flash
Shipped Sep 2; stronger docs/video/retrieval.
View details
New capabilities
- 3.8 Flash on 2 September
- Stronger docs, long video, and retrieval
What got stronger
- Smoother on sheets, slides, and recordings than chat-only models
Version notes
3.8 Flash continues fast, cheap, and fluent with Google files. If your work lives in Workspace, that beats a better poet. For coding agents, still compare Claude / GPT flagships.
Aug 2026
Gemini 3.6 / 3.7 Flash
Rapid Flash cadence; lower cost.
View details
New capabilities
- 3.6 / 3.7 Flash on a near-weekly cadence
- Kept pushing inference cost down
What got stronger
- Quality quietly improved more often than rivals
Version notes
In summer 2026 Google used rapid Flash drops for data and share. Users felt Gemini get better at docs and sheets more than a sudden coding leap.
Mar 2025
Gemini 2.5 Pro
Reasoning and long-context flagship.
View details
New capabilities
- 2.5 Pro as reasoning + long-context flagship
- Deeper multi-step planning
What got stronger
- Closer to then-frontier closed models on hard and long tasks
Version notes
2.5 Pro was the serious flagship after the Flash rush. Coding still often trailed Claude / GPT; documents, sheets, video, and retrieval were home turf.
Dec 2024
Gemini 2.0 Flash
Stronger agents and native tools.
View details
New capabilities
- Native tool use and agent loops
- Flash: fast, cheap default tier
What got stronger
- Latency and price fit Search, Gmail, and IDEs
- Agents moved beyond demo videos
Version notes
2.0 Flash moved Gemini from “can read a lot” to “can act.” The Flash tier made it financially plausible to embed in almost every Google product.
Feb 2024
Gemini 1.5 Pro
Million-token context experiments.
View details
New capabilities
- Million-token context experiments
- Whole codebases or video transcripts
What got stronger
- Context window size became Google’s headline
- Better for “stuff it all in, then ask” research
Version notes
1.5 Pro made million-token context a demo-grade feature: hour-long video, full PDFs, mid-size repos. Quotas and stability were still climbing, but the strategy was clear: win on window size.
Dec 2023
Gemini 1.0
Bard becomes Gemini; native multimodal.
View details
New capabilities
- Bard rebranded to Gemini
- Native multimodal: image and text trained together
- Hooked to Google accounts and early Workspace
What got stronger
- No longer a chat skin over PaLM
- Search, Photos, and YouTube started to differentiate it
Version notes
Gemini 1.0 was Google naming its own flagship assistant. Native multimodal was the pitch; early polish was mixed. The real gap came later with huge context and Google-surface distribution.