Grok vs Gemini — which actually wins in 2026?
This pair splits cleanly. Grok 4.7 leads the composite — Intelligence Index 46.5 vs 40.9 — and terminal work, 25.8% vs 19.7% on Terminal-Bench 4.0. Gemini 3.8 Flash costs a fraction of it, $0.75 / $3.75 against $2.00 / $6.00, takes audio and video, and doubles the context. Google has published a price rise for January 1, 2027.
| Category | Winner | Margin |
|---|---|---|
| Composite intelligence · AA Intelligence Index v4.3.2 | AGrok | Grok 4.7 46.5 vs Gemini 3.8 Flash 40.9 (Artificial Analysis, read 2026-09-23; Grok at xhigh, Gemini at high) |
| Agentic terminal work · Terminal-Bench 4.0 | AGrok | 25.8% vs 19.7% on AA's uniform harness, and 37.6% vs 19.1% on tbench.ai, where Grok runs through Grok Build and Gemini through mini-SWE-agent |
| Hard-exam reasoning · Humanity's Last Exam | BGemini | 47.8% vs 43.1% to Gemini (Artificial Analysis, 2026-09-23) |
| Real-time information · live sources | AGrok | Grok reads live X posts inline; Gemini grounds on Google Search, billed at $14 per 1,000 requests after 5,000 free a month |
| API context window · API maximum | BGemini | 1M tokens against Grok's 500K, and no doubling step — Grok charges double above a 200K prompt |
| Multimodal input · what you can send | BGemini | Gemini 3.8 Flash accepts audio and video as well as text and images; Grok 4.7 takes text and images only |
| API cost · list rates per 1M tokens | BGemini | $0.75 / $3.75 against $2.00 / $6.00 — Gemini is cheaper on both, and blends to $0.481 against $1.19 |
| Published price risk · announced changes | AGrok | Google has published a doubling of Gemini 3.8 Flash's rates from January 1, 2027; SpaceXAI has announced no change to Grok 4.7's card |
| Batch discount · async work | BGemini | Gemini's Batch and Flex tiers are half standard ($0.375 / $1.88); Grok 4.7 publishes no batch discount |
| Free tier · what $0 buys | BGemini | Gemini's free app tier runs 3.6 Flash with varying access to 3.1 Pro, and the API has a free tier whose content Google uses to improve its products |
| Office integration · where it lives | BGemini | Gemini is native in Gmail, Docs, Vids and Search; Grok offers connectors, the Grok Build CLI and X |
| Entry subscription · cheapest paid tier | BGemini | Google AI Plus is $4.99/mo and AI Pro $19.99/mo; the cheapest standalone Grok tier is SuperGrok at $30/mo |
| Best overall · for most teams | ·Depends | See the decision tree below |
If you need real-time and cheap output tokens.
- Composite lead — Grok 4.7 tops Artificial Analysis' Intelligence Index v4.3.2, 46.5 against 40.9
- Terminal work — 25.8% against 19.7% on AA's harness, 37.6% against 19.1% on tbench.ai
- Real-time X — live posts read inline, which Google has no equivalent of
- Media in-thread — Grok Imagine generates images and video inside the same chat
- Open harness — Grok Build is Apache-2.0 and ships on every plan, Free included
If you need context, multimodality and Google-native work.
- 1M context — double Grok 4.7's window, and with no doubling step above 200K
- Audio and video in — Gemini 3.8 Flash ingests both natively; Grok 4.7 takes text and images
- Cost — $0.75 / $3.75 against $2.00 / $6.00, and half that again on Batch
- Where you already work — Gmail, Docs, Vids, Search and the Workspace stack
- Cheap entry — Google AI Plus at $4.99/mo, AI Pro at $19.99/mo
| Aspect | Grok | Gemini |
|---|---|---|
| Free tierwhat $0 buys | $0 · Grok Free Real-time web and X search, voice mode, connectors and Grok Build, inside limits SpaceXAI does not quantify | $0 · Gemini Free Access to 3.6 Flash with varying access to 3.1 Pro, image generation, Deep Research, Gemini Live and Canvas, plus 15 GB of storage B wins |
| Entry subscriptioncheapest paid tier | $30/mo · SuperGrok Higher limits across Chat, Imagine, Voice and Build; the pricing page still names Grok 4.6 as the model it unlocks | $4.99/mo · Google AI Plus 2x free-tier usage, video generation, Daily Brief, 200 Google Flow credits and 400 GB of storage B wins |
| Mainstream subscriptionthe tier most people buy | $100/mo · SuperGrok Plus Significantly higher usage, 1080p video, priority access at peak times | $19.99/mo · Google AI Pro 4x free-tier usage, 1,000 Flow credits, and access to Gemini 3.1 Pro in the app B wins |
| Power tierfor all-day use | $300/mo · SuperGrok Heavy Top individual Grok tier; its price is published on the Grok Bot page rather than the pricing page | $99.99–199.99/mo · AI Ultra $99.99 for 5x AI Pro usage, $199.99 for 20x, plus first access to features like Deep Think B wins |
| API · inputper 1M tokens · from snapshot | $2.00 | $0.75 B wins |
| API · outputper 1M tokens · from snapshot | $6.00 | $3.75 B wins |
| Effective API costBlended workload $/1M · from snapshot | $1.19 | $0.481 B wins |
| API context windowMax input tokens · from snapshot | 500K | 1M B wins |
| Real cost / 1M charsTokenizer-adjusted prose — the tokenizer tax | $0.519 | $0.195 B wins |
| Team / enterpriseper seat | $30/seat · Grok Business Seat management, RBAC, SSO, no training on business content | Bundled · Workspace / Enterprise Gemini rides Google Workspace plans and the Gemini Enterprise Agent Platform; gemini.google prints consumer tiers only, so we quote no seat price here |
| Capability | Grok | Gemini |
|---|---|---|
| API context window | 500K tokens | 1M tokens |
| Max output tokens | ~ Not published | 65K tokens |
| Vision / image input | ✓ Images, screenshots | ✓ Images, PDFs, screenshots |
| Audio input | ✗ | ✓ Native |
| Video input | ✗ | ✓ Native |
| Image generation | ✓ Grok Imagine | ✓ In-app image gen |
| Video generation | ✓ Grok Imagine | ✓ Veo (paid tiers) |
| Voice mode | ✓ Voice mode (free too) | ✓ Gemini Live |
| Real-time X integration | ✓ Native | ✗ (Google Search only) |
| Search grounding | ✓ Live web + X | ✓ Native Google Search |
| Deep research runs | ~ Not a named feature | ✓ Deep Research (free tier too) |
| Code execution sandbox | ✓ Grok Build | ✓ Canvas / code exec |
| Coding agent | ✓ Grok Build, Apache-2.0, on every plan | ~ Gemini CLI and third-party harnesses |
| Custom assistants | ✗ | ~ Gems (no public store) |
| Connectors | ✓ Connectors (free too) | ✓ Workspace + Drive |
| Office / productivity integration | ~ Connectors only | ✓ Native Workspace |
| Function calling / tool use | ✓ Native | ✓ Native |
| Prompt caching | ✓ $0.50/M read | ✓ $0.075/M read, plus storage per hour |
| Batch API discount | ✗ (not published for Grok 4.7) | ✓ $0.375 / $1.88 per 1M |
| SOC 2 compliance | ✓ Type I & II (all tiers) | ✓ Workspace / Cloud |
| SSO / SCIM | ✓ Business tier | ✓ Workspace admin |
| Desktop apps | ✗ (web only) | ~ Web + Chrome |
| Mobile apps | iOS + Android | iOS + Android |
| EU availability | ~ US API regions only; resold on Bedrock and Foundry | ✓ Broad |
The numbers, not the spin.
Grok
The real-time-first assistant — native to X, ahead on the composite index, and cheaper where output tokens dominate the bill.
Strengths
- Composite lead — 46.5 against 40.9 on AA's Intelligence Index v4.3.2
- Terminal work — ahead on both Terminal-Bench 4.0 boards, by 6 points on AA's harness and 18 on tbench.ai
- Real-time X — live posts inline, billed per post fetched
- Media — images and video generated in-thread by Grok Imagine
- Open harness — Grok Build under Apache-2.0 on every plan
Weaknesses
- Costs roughly 2.7x Gemini on input and 1.6x on output, before cache
- No audio or video input; Grok 4.7 is text + image only
- 500K context, and every rate doubles at a 200K-token prompt
- No batch discount, where Gemini halves both rates
Best for
- X-native creators, journalists and analysts
- Real-time event coverage and breaking-news research
- Output-heavy API workloads where generation dominates spend
- Prompts other assistants deflect
Gemini
The context-and-Google-native assistant — 1M tokens, audio and video input, Workspace integration, and the broader free tier of the two.
Strengths
- 1M context — double Grok 4.7's window on the API model
- Audio and video input — native, on the same rate card
- Cheapest card in this pair — $0.75 / $3.75, halved again on Batch and Flex
- Hard-exam reasoning — 47.8% on Humanity's Last Exam against Grok's 43.1%
- Distribution — Gmail, Docs, Vids, Search and Workspace
Weaknesses
- Behind on the composite index and on both Terminal-Bench boards
- Google has published a doubling of these rates from January 1, 2027
- Search grounding is billed separately: $14 per 1,000 requests after 5,000 free a month
- The free API tier is explicitly used to improve Google's products
Best for
- Long-document and long-context work
- Teams already living in Google Workspace
- Audio and video analysis pipelines
- Prompt-heavy API workloads that lean on caching and batch
- Anyone in the EU
Journalist covering breaking news on X
You file several stories a day during fast-moving events and need to verify quotes and pull live reactions from X within minutes.
Reasoning: Grok reads live X posts inline, so you skip the copy-paste from a separate tab. Gemini grounds on Google Search with citations, which is better for the wider web but has no X-native path. For X-driven stories, Grok cuts the most time out of the loop.
Analyst summarising a 400-page filing
You drop long PDFs and transcripts into an assistant and ask it to trace claims across the whole document rather than a chunked subset.
Reasoning: Gemini 3.8 Flash carries a 1M-token window against Grok 4.7's 500K, so more of the document survives in a single pass — and the last 300K of Grok's window costs double, because every Grok rate doubles once a prompt reaches 200K tokens. Gemini also takes audio and video natively, which matters when the "document" is a recording. The counterweight is the dated one: Google has published a doubling of Gemini 3.8 Flash's rates from January 1, 2027, so a long-lived pipeline should price both years.
API developer on a generation-heavy workload
Your product streams long completions — drafts, summaries, agent turns — so output tokens, not prompts, dominate the monthly bill.
Reasoning: Gemini now undercuts Grok on every axis: $0.75 vs $2 input, $3.75 vs $6 output, $0.075 vs $0.30 cached. Google halved the Flash card on August 14, 2026, which reversed what used to be Grok's output-price advantage. The one caveat is duration — Gemini's rate is scheduled to double on January 1, 2027, while Grok's is flat. The verdict flips on prompt-heavy workloads. Check the blended and tokenizer-adjusted rows above against your own input/output mix.
Team that lives in Google Workspace
Your day is Gmail, Docs, Sheets and Meet, and you want an assistant that reads and writes across those without copy-paste.
Reasoning: Gemini is native inside Workspace and bundled at no extra AI charge, so the marginal cost of rolling it out to a team already paying for Workspace is zero. Grok offers connectors but no Office-class integration, and it publishes no seat price for its Business tier. For a Google-first team, this isn't close.
Developer or user based in the EU
You're building from, or working in, the EU and need the model available through a normal console and account today.
Reasoning: Region terms decide this one. SpaceXAI serves Grok from US regions only (us-east-1, us-west-2, us-central-1), with a US regional endpoint at a 10% premium and no EU processing commitment; EU users did get API-console access to Grok 4.5 in July 2026, which is availability rather than residency. Google runs Gemini broadly and sells enterprise terms through Workspace and the Gemini Enterprise Agent Platform. If EU data residency is contractual for you, that asymmetry is the whole decision.
Researcher on a zero budget
You want serious assistant work — research runs, image generation, a canvas to iterate in — without paying anything.
Reasoning: Gemini's free app tier runs 3.6 Flash with varying access to 3.1 Pro, plus image generation, Deep Research, Gemini Live and Canvas, and 15 GB of storage. Grok's free tier gives real-time web and X search, voice mode, connectors and Grok Build, but SpaceXAI does not quantify the limits anywhere. One free tier is described in features, the other in adjectives — and on the API side Google's free tier is explicit that content is used to improve its products.
Frequently asked.
Common questions about this comparison, with sources where they matter.
Q · 01 Is Grok or Gemini better overall? +
54 vs 50), and pricing that carries no expiry date ($6/M output flat, against Gemini's promotional $7.50/M). Gemini leads on context (1M vs 500K), audio and video input, input price, Workspace integration, free-tier generosity, and EU availability. On GPQA Diamond they tie (93.1 vs 94.1, inside the error band). The best pick depends on your workload — see the decision tree above.Q · 02 Which is cheaper? +
$0.75 against $2.00, output $3.75 against $6.00, cache reads $0.075 against $0.50, and Batch at half standard where Grok 4.7 has no batch discount at all. Blended on our standard agentic mix that is $0.481 against $1.19. One caveat with a date on it: Google has published that these rates double on January 1, 2027.Q · 03 Which has the larger context window? +
1M-token window against Grok 4.7's 500K, and Google charges one rate across the whole window. Grok charges two: at or above a 200K-token prompt every rate doubles, for the entire request, so the upper half of Grok's smaller window is also its expensive half.Q · 04 Which is better for coding? +
25.8% and Gemini 3.8 Flash at 19.7%, and the benchmark's own board at tbench.ai — where Grok runs through Grok Build and Gemini through mini-SWE-agent — has 37.6% against 19.1%. Both boards agree on the direction, which is more than the July version of this page could say. Neither model has a neutral SWE-bench Verified entry.