Claude vs Gemini in August 2026

Claude and Google Gemini compared on the things that differ most - long-context retrieval, speed, multimodal generation and price - with verified August 2026 figures.

Claude and Gemini are not competing on the same axis. As of August 2026, Google's flagship is Gemini 3.7 Flash - released 13 August 2026 - and it beats Claude decisively on long-context retrieval, raw speed and price, while Anthropic's Opus 5 and Fable 5 hold the top of the independent intelligence and human-preference rankings. Google also generates images, video and music. Claude generates none.

The two clearest Google wins come first on this page, because they are the ones most comparison articles bury.

#Where does Gemini beat Claude?

#1. Long-context retrieval - a 1M window is not the same as reliable recall

This is the most important finding on the page, and it cuts against the marketing on both sides. Claude offers a 1,000,000-token context window as the default on Fable 5, Opus 5 and Sonnet 5, with no beta header and no long-context price premium. That is a genuine engineering achievement. But window size measures what you can load, not what the model can reliably find again.

On GDM-MRCR v2, a multi-round needle-retrieval benchmark, Google's own Gemini 3.7 Flash model card reports 97.0% against Claude Sonnet 5's 81.5% and GPT-5.6 Terra's 93.5%. On LVBench, a long-video benchmark, it is 85.4% against Sonnet 5's 68.5% - though that comparison is partly moot, since Claude cannot accept video input at all. Claude has published very little long-context retrieval evidence of its own, and loses the retrieval benchmarks that do exist. The one exception in the published record runs the other way: on GraphWalks BFS at 1M tokens, OpenAI's launch table gives Anthropic's Mythos 5 79.4% against GPT-5.6 Sol's 77.1%.

Practical reading: if your workload is "drop 600 pages in and ask precise questions about page 412", Gemini is the safer tool. If it is "reason carefully across a codebase you have structured yourself", the retrieval gap matters much less. Our model comparison works through which Claude model to use when context length is the binding constraint.

#2. Speed and price - Google wins by an order of magnitude

Artificial Analysis measures Gemini 3.7 Flash at roughly 3,901 output tokens per second. Fable 5 runs at roughly 75. That is not a tuning difference; it is a different class of product. Opus 5 is much better at around 626 tokens/sec, but still six times slower.

Gemini 3.7 Flash's $0.75 in / $3.75 out is introductory pricing through 31 December 2026, and Google's Gemini API pricing page states rates double to $1.50/$7.50 on 1 January 2027. Even after that doubling it undercuts Sonnet 5 at $2/$10. Anthropic's counter-levers are the Batch API's flat 50% discount and prompt caching at 0.1× base input on cache reads - both covered on the Claude API page - but they apply to Google's rates too.

#3. Hard-reasoning breadth

On Humanity's Last Exam (Verified), Google's card reports Gemini 3.7 Flash at 53.6% against Claude Sonnet 5's 31.0% - a 22-point gap against a model costing roughly a quarter as much. Anthropic has not published an Opus 5 or Fable 5 figure on that benchmark, so the flagship-to-flagship comparison does not exist. Treat the 22 points as a mid-tier result, not a settled verdict on the whole family, but note that the absence of an Anthropic number is itself part of the problem.

#4. Generative media, which Claude simply does not do

Google ships Veo 3.1 for video, Imagen for images, Lyria 3 for music, Flow and Flow Music for production workflows, and Gemini Omni for multimodal creation. Gemini 3.7 Flash accepts text, image, audio and video as input. Claude accepts text and images and emits text only - no image, video, audio or music generation, and no video or audio input. For any creative or media pipeline this is decisive, and no benchmark changes it. The same gap against OpenAI is covered on our ChatGPT comparison.

#5. Distribution

Gemini is the default assistant in Google Search's AI Mode, in Gmail, in Docs and on Android, plus NotebookLM and a bundled consumer stack. Anthropic has no search engine, no mobile OS and no office suite. Claude reaches you because you went looking for it; Gemini reaches you because you opened your inbox.

#What is each company actually shipping?

One structural fact is worth pausing on: Google's best model is a Flash model. The Pro line has not shipped a stable release past Gemini 3.1 Pro, which remains in preview. Google describes 3.7 Flash as its "most intelligent workhorse model yet for coding and agents" - that is a company competing on intelligence-per-dollar, not on an absolute frontier model. Anthropic is doing the opposite.

Gemini 3.7 Flash's knowledge cutoff is March 2026 (January 2025 in some domains). Among Claude models, Opus 5 has the newest at May 2026 - cutoffs are not in version order, since Fable 5 and Sonnet 5 both stop at January 2026. The full specification set is on the Claude models index.

#Where does Claude beat Gemini?

Four areas, and they are not consolation prizes.

Two things make that table trustworthy. First, several of those Claude wins appear on Google's own model card - Google did not cherry-pick a sweep, which is to its credit. Second, the Intelligence Index and Elo figures come from third parties running their own harness rather than from either vendor.

But read Google's card carefully in the other direction too: it compares Gemini 3.7 Flash against Sonnet 5 and GPT-5.6 Terra, not against Fable 5 or Opus 5. Gemini beats Sonnet 5 on twelve of eighteen rows at roughly a quarter of the price, which is a real result - but it is a mid-tier comparison being read as a flagship one. Anthropic makes the mirror-image mistake: its Fable 5 benchmark table is published only as an image, and the Opus 5 post gives relative claims rather than absolute numbers, so the flagship-to-flagship table nobody can build is partly Anthropic's fault.

#Agentic coding is still Claude's strongest ground

Claude Code plus Fable 5 leads the official Terminal-Bench 2.1 board at 83.8%; no Gemini model appears near the top of Terminal-Bench or SWE-bench. Where Gemini does compete on code, it competes on value - it edges Sonnet 5 on FrontierCode 1.1 (43.6% vs 42.7%) and leads Code Arena Web Elo (1588 vs 1541) at a quarter of the cost. That is a strong argument for high-volume code work and a weak one for the hardest single task in your backlog. The tooling side is covered on Claude for developers.

#Enterprise compliance depth

Anthropic publishes Enterprise at a flat $20/seat/month plus usage at API rates, with SCIM, audit logs, a Compliance API, custom data retention, customer-managed encryption keys, US-only inference, IP allowlisting and HIPAA-ready configuration listed openly. Google's strength is different in kind: it wins where the customer already lives in Workspace and GCP, with Gemini Enterprise Agent Platform as the deployment path. If your procurement process is a control checklist rather than an existing contract, Anthropic's published surface is easier to evaluate.

#How do the consumer plans compare?

Claude plan prices are from claude.com/pricing and explained in detail on the pricing page; what the free tier actually includes is on the Claude free plan page. Note that Google's $19.99 tier ships generative video and music inside the subscription, which Claude Pro cannot match at any price.

#Claude or Gemini: which should you pick?

If neither answer fits, the wider field is surveyed on Claude alternatives - including open-weight models now within a few Intelligence Index points of the frontier. Our overall verdict on whether Claude is worth its price is on the Claude review page, and the OpenAI comparison is on Claude vs ChatGPT.

#Frequently asked questions