Gemini 3.1 Pro is Google's most credible frontier offering yet, closer on coding and reasoning than the 2.x line, and still ahead of everyone on context size (1M baseline, 2M extended). Multimodal video understanding remains best-in-class. The trade is the same one it's always been: deeper Google ecosystem coupling, pricing tier complexity, and a coding gap to Claude that's narrowed but hasn't closed. Default choice if you're on Google Cloud or your workloads are video/long-document heavy.
- 1M-token baseline context, 2M extended, largest in the frontier
- Best-in-class multimodal video understanding
- Deep Research, Computer Use, and Nano Banana 2 image gen
- Tight Google Cloud + Workspace integration
- Gemini 3 Flash and Flash-Lite for cost-efficient production
- Pricing tier jumps at 200K, surcharge above doubles input cost
- Coding scores trail Claude on most current benchmarks
- Free tier rate limits tightened in 2026
- Vertex AI vs Gemini API split adds onboarding friction
- Best features land on Google Cloud first, third-party later
Gemini is Google DeepMind’s frontier model family. It ships as the
Gemini app and Workspace integrations (consumer, free + Google AI
Pro $20/mo + Ultra), the Gemini API on ai.google.dev, and Vertex
AI for enterprise deployments. The 3.x line in 2026 closed most of
the credibility gap that the 2.x line still carried.
Where it fits
Gemini is the right pick when long context or multimodal work dominates. The 1M-token baseline window with 2M extended is the largest among frontier proprietary models, which materially changes what “load the whole repo into the prompt” looks like. Native multimodal, text, image, audio, video, PDF, code, is built in rather than bolted on, and video understanding specifically remains ahead of both Claude and GPT.
Outside those two wedges, Gemini’s positioning is “viable alternative” rather than category lead. Coding scores trail Claude on most published benchmarks (though the gap has narrowed from the 2.x line). For pure text/coding work without the context or multimodal pull, Claude or GPT are usually the default.
Pricing in practice
API pricing (per 1M tokens): Gemini 3.1 Pro at $2/$12 input/output
for prompts up to 200K. Above 200K, pricing jumps to $4/$18, the
context surcharge is a real cost lever to model around. Gemini 3
Flash and Gemini 3.1 Flash-Lite are the budget tiers for
high-volume work. For consumer use, the free tier on gemini.google.com
covers light use; Google AI Pro is $20/mo with deeper limits and
Ultra ($250/mo) bundles Veo video generation and longer Deep
Research runs. Vertex AI on Google Cloud is the enterprise route
with the same per-token rates but integrated billing.
How it compares
Claude, Stronger on coding and agent workflows at comparable price. Pick when coding or agentic work is the centerpiece.
GPT, Broader third-party integration, better image generation, larger consumer footprint. Pick when ecosystem reach matters.
DeepSeek, Open-weight V4 Pro near-frontier at much lower cost. Pick when cost or self-hosting is the constraint.
Llama, Open weights, run anywhere. Behind the frontier on benchmarks. Pick when self-hosting is non-negotiable.
Latest news
Google launched Gemini 3.5 on 2026-05-19 at I/O 2026, framed as “frontier intelligence with action”, positioning the release around agentic, tool-using deployments rather than chat-first use. The drop succeeds the Gemini 3.1 line that has been the recommended default since February 2026. Gemini 3.5 Flash reached hosted-inference partners the same week, appearing on the Vercel AI Gateway (2026-05-19) and on Netlify’s AI Gateway and Agent Runners (2026-05-19/20). I/O 2026 itself was a wide-surface event, Google’s full announcement digest covered roughly 100 separate items spanning Workspace, Android, and Cloud, with Gemini 3.5 the headline platform release.
Sources
- Gemini API models, ai.google.dev, May 2026
- Gemini Developer API pricing, ai.google.dev
- Gemini 3.1 Pro API pricing 2026, glbgpt.com
- Google Gemini API Pricing May 2026, aipricing.guru
- Gemini Pricing 2026, felloai.com
- launch · 2026-08-14
Gemini 3.7 Flash lands three weeks after 3.6, with a real coding jump and a half-price countdown
Google shipped Gemini 3.7 Flash three weeks after 3.6, pairing real gains on agentic-coding benchmarks with an introductory price set at half the model it replaces. The catch: that rate doubles back to the old level on January 1, 2027.
- beat · 2026-08-08
OpenAI pauses Astra, the first model to trip its own critical-cyber threshold
OpenAI paused internal work on Astra after concluding it may cross the "critical" cyber bar in its Preparedness Framework: able to find and exploit zero-days on hardened systems without human help. It is the first time that gate has stopped a model.
- follow-up · 2026-08-03
OpenAI cuts GPT-5.6 Luna 80% three weeks after launch, and the budget tier becomes the battleground
OpenAI cut GPT-5.6 Luna 80% on July 30, three weeks after launch, dropping its budget tier to $0.20 and $1.20 per million tokens while leaving the Sol flagship untouched. The move points the frontier price war at the tier where most production traffic actually runs.
- explainer · 2026-08-02
The Model Context Protocol goes stateless, and starts a deprecation clock on Roots and Sampling
MCP's July 28 revision removes protocol sessions and the initialize handshake, treats servers as OAuth 2.1 resource servers, and starts a twelve-month clock on Roots, Sampling, and Logging. Vercel and WorkOS already ship support.
- launch · 2026-07-24
Claude Opus 5 holds Opus pricing flat while chasing Fable 5's coding lead
Anthropic released Claude Opus 5 on July 24 and priced it exactly where Opus 4.8 sat, $5 and $25 per million tokens, while claiming a smaller model that trades blows with the current frontier on agentic coding. The race keeps shifting from capability to cost per completed task.