Claude vs Gemini 2026: Who Gives You 1M Context Free?

For most of 2026 there was a clean line you could use to end this argument: Gemini has the bigger window, Claude has the better answer. That line has just stopped being true in one direction and become more true in the other, and the reason is a documentation change rather than a launch.

Anthropic's help centre now lists a 1M token context window on paid Claude plans. It did not say that three months ago. What it said, and what we repeated on this site as recently as our last comparison, was that every consumer tier shared a 200K window and that paying bought usage rather than room. Paying now buys both. The free plan did not come along for the ride.

On the Google side, the newest Flash model went generally available in early September, but it landed on the API rather than in the free app. So the free Gemini experience is unchanged, which in this particular argument is the point: it did not need to change.

Which Is Better in 2026, Claude or Gemini?

Gemini for free users. Claude for paying users doing hard language work. That split is cleaner now than it has been all year, and it is worth being blunt about why.

If you are not paying, this is close to a rout. Free Gemini hands you a 1M token window on a Flash-tier model along with image generation, web search, code execution and Workspace integration. Free Claude hands you 200K tokens on Sonnet and Haiku, at half of Pro's weekly allowance, with Claude Code, the Opus model and Projects all sitting behind the paywall according to Anthropic's own pricing page. Five times the context and more features, for the same amount of money, which is none.

If you are paying, the calculation inverts because the thing Gemini was winning on is no longer a Claude weakness. A Claude Pro subscriber now gets the same 1M window, and Claude still writes better prose, holds a voice across a long piece and follows layered instructions more precisely. You are paying for the quality of each answer rather than for room to put documents in.

What we are not going to tell you is that it depends on your use case. It depends on whether you are paying, which is a question you can answer in one second.

What Actually Changed This Month?

Two things, one on each side, and neither was announced with a launch event.

Anthropic split the context window by tier. The Claude help centre article on paid plan context windows documents 1M tokens in chat for Claude Fable 5.1, Claude Opus 5 and Claude Sonnet 5, 500K for Opus 4.6 through 4.8 and Sonnet 4.6, and 200K for everything else. It also notes that in Claude Cowork, Sonnet 5 automatically compacts the conversation at 500K tokens, which is a detail worth knowing before you plan a very long session. The article is scoped to paid plans and does not cover the free tier, which remains at 200K.

That correction changed two pages on this site. Our Claude review previously said all three consumer tiers shared a 200K window and that upgrading bought usage rather than room. It said so because that was accurate when we wrote it. It is not accurate now, and the page has been rewritten.

Google shipped Gemini 3.8 Flash to the API. The Gemini API changelog records general availability on 2 September 2026 as gemini-3.8-flash, described as Google's most intelligent Flash model, built for long-horizon software engineering, autonomous agents and complex enterprise workflows. The model page puts it at a 1M token context window with 64K max output and tunable thinking levels.

Here is the part most roundups skip. That is an API release. Google's Gemini app release notes do not document 3.8 Flash becoming the default for free app users, and the free tier continues to run a Flash-tier model. If you read a headline saying you can now use Gemini 3.8 Flash for free in the app, check whether the writer distinguished the API from the product. Most did not.

How Do the Context Windows Compare Now?

A tie on paid, a five to one win for Gemini on free.

On paid plans both sit at 1M tokens in the chat interface. Claude Fable 5.1, Opus 5 and Sonnet 5 are documented at 1M, and Gemini 3.8 Flash is documented at 1M. At that size the number stops being the interesting variable, because almost nothing an individual brings to a chat window comes close to filling it. A 1M window is roughly 555,000 words on Anthropic's current tokenizer, which is several full-length books.

On free plans the comparison is stark. Gemini gives 1M. Claude gives 200K, or around 150,000 words. For most work that 200K is genuinely enough: a contract, a report, a thesis chapter, a codebase module. The people it fails are the ones who paste in a whole document set rather than a document, and that is a real population rather than an edge case.

One caveat that cuts against Gemini and deserves saying. A large window is a ceiling, not a promise about attention. A model that accepts a million tokens and skims them is worse for contract review than one that accepts 200,000 and reads all of them. We have not found a way to test attention quality across a full window rigorously enough to publish a number, so treat the context comparison as a capacity comparison and nothing more.

Which One Wins on the Free Tier?

Gemini, decisively, and this is the section most readers came for.

Free Gemini gives you a Flash-tier model with a 1M token window, image generation, web search, file uploads, code execution and Google Workspace integration, on limits generous enough that ordinary daily use rarely touches them. It asks for a Google account and nothing else.

Free Claude gives you Sonnet and Haiku at 50% of Pro weekly limits inside a 200K window, with web search, file uploads, file creation and code execution. Anthropic's pricing page is explicit that free users do not get Claude Code, do not get the Opus model and do not get Projects. The weekly limit is a real ceiling and long document sessions reach it faster than people expect, because the conversations Claude is best at are exactly the expensive ones.

So the honest framing is not that Claude's free tier is bad. It is that Claude's free tier is a sample of a paid product, and Gemini's free tier is a product. If your question is which one to open when you are not paying anything, open Gemini and keep Claude for the handful of tasks where the answer quality genuinely decides something. For a wider view of what else is free right now, our sister site maintains a verified list of genuinely free AI tools checked against vendor documentation rather than press releases.

What Do They Cost to Run on the API?

Gemini, by a wide margin, and the margin survives the introductory period ending.

Gemini 3.8 Flash is priced at $0.75 per million input tokens and $3.75 per million output as an introductory rate through 31 December 2026. On 1 January 2027 that becomes $1.50 and $7.50. Both figures come from Google's own model page rather than a third party tracker.

On the Claude side, the models overview lists Sonnet 5 at $2 input and $10 output, Opus 5 at $5 and $25, and Fable 5.1 at $10 and $50. Batch requests are half price and cache reads cost 10% of base input, dropping to 2.5% on Fable 5.1.

Put plainly: Gemini 3.8 Flash at its introductory rate is cheaper on input than Claude Sonnet 5 by a factor of roughly two and a half, and cheaper than Fable 5.1 by roughly thirteen. Even at the January 2027 standard rate it stays below Sonnet 5. These are not the same class of model and pretending otherwise would be dishonest, but for high volume classification, extraction and summarisation the quality difference often does not show up in the output while the price difference shows up on every invoice.

The caveat that stops this being a rout is caching. A Claude workload with a large stable prefix and heavy reuse pays 2.5% of base input on Fable 5.1 cache reads, which closes a lot of the gap for agent architectures that resend the same system context hundreds of times a day.

Which Is Better for Writing and Long Documents?

Claude, and the newer context numbers have not changed that.

Claude produces more natural prose, keeps a consistent voice across a long piece, and follows multi-part instructions without quietly dropping the third one. On document work it reads more carefully rather than merely reading more. Gemini is faster and more willing to be comprehensive, which is the right trade for research summaries and the wrong one for a piece someone will read closely.

Gemini's real advantage in this category is not the model, it is the plumbing. It reads your Google Drive, drafts inside Docs and pulls from Gmail without an upload step. For anyone who already lives in Workspace, that removes a friction that no amount of prose quality compensates for.

The honest weakness on the Claude side, for free users specifically: if your document exceeds 200K tokens you cannot use Claude free for it at all, and no writing quality fixes a hard limit. Check the size before you choose the model.

Output quality on both is more sensitive to how you ask than to which one you picked, which is the least fashionable thing to say in a comparison article and the most consistently true. Structured prompt patterns for writing and analysis tasks are collected at PromptCraft Asia, and our full Claude vs Gemini comparison goes task by task.

Do Both Work Properly From Southeast Asia?

Yes, both, no VPN, across Singapore, Malaysia, the Philippines, Indonesia and Thailand.

Neither vendor gates access regionally and both accept Singapore-issued cards on paid plans. Claude Pro is $20 a month billed monthly or $17 a month billed annually, which is roughly SGD 27 and SGD 23, with Claude Max starting at $100 a month.

The regional differentiator is language rather than access. Gemini handles Malay and Thai noticeably better than Claude does, and Google's training data advantage in regional languages is not something Anthropic has closed. If you work in English, ignore this. If you draft in Bahasa Malaysia or Thai, it outweighs most of the other differences in this article.

On data residency both store conversations in the United States. For ordinary work that is unremarkable. For regulated material it is the question you answer before any of the numbers above matter, and it is why both of these remain easier to justify for business use than the cheaper China-hosted alternatives regardless of price.

Which Should You Pick for Your Own Work?

Four rules that cover nearly everyone.

Pick free Gemini if you are not paying. A 1M window, image generation and code execution at no cost beats a 200K sample of a better model for almost every practical purpose.

Pick Claude Pro if you are paying and your work is writing, editing, analysis or careful reasoning. You now get the same 1M window as Gemini, so the only question left is answer quality, and Claude wins that one.

Pick Gemini on the API if you are building something with volume. Gemini 3.8 Flash undercuts every frontier Claude model on price, and for extraction and summarisation the output gap is usually smaller than the invoice gap.

Pick Claude on the API if the workload has a large stable prefix you resend constantly, because cache reads at 2.5% of base input on Fable 5.1 change the arithmetic substantially, or if the task is long-horizon agentic work that Anthropic built Fable 5.1 for specifically.

And if you are still torn, run one genuine task through both this week. Two hours of your own testing beats any comparison table, this one included.

What Else Do People Ask?

Does Claude have a 1M token context window now?

On paid plans, yes. Anthropic's help centre documents a 1M token context window in chat for Claude Fable 5.1, Claude Opus 5 and Claude Sonnet 5 on paid plans, with the older Opus 4.6 through 4.8 and Sonnet 4.6 at 500K and everything else at 200K. The free plan is not included in that documentation and still runs a 200K window. This is a change from earlier in 2026, when every consumer tier shared 200K and paying bought usage rather than room. It now buys both, which makes the free versus paid decision a different calculation than it was three months ago.

Is Gemini 3.8 Flash available on the free Gemini app?

Not as the free default. Google's API changelog records Gemini 3.8 Flash reaching general availability on 2 September 2026 as gemini-3.8-flash, described as its most intelligent Flash model for long-horizon software engineering and autonomous agents. That is an API release. Google's Gemini app release notes do not document 3.8 Flash arriving as the default model for free app users, and the free tier continues to serve a Flash-tier model. The practical point for free users is that the context window is 1M either way, so the model version matters more for coding and agent work than for everyday questions.

Which has the better free tier, Claude or Gemini?

Gemini, and the gap widened this quarter. Free Gemini gives a 1M token context window, image generation, web search, code execution and Google Workspace integration on a Flash-tier model. Free Claude gives a 200K window on Sonnet and Haiku at 50% of Pro weekly limits, and Anthropic's pricing page lists Claude Code, the Opus model and Projects as paid-only. So Gemini offers five times the context, more features and no weekly ceiling worth worrying about. Claude free still answers better on hard writing and reasoning, which is the only reason to pick it.

How much do Claude and Gemini cost per million tokens?

Gemini 3.8 Flash runs at an introductory $0.75 per million input tokens and $3.75 per million output through 31 December 2026, rising to $1.50 and $7.50 on 1 January 2027. On the Claude side, Sonnet 5 costs $2 input and $10 output, Opus 5 costs $5 and $25, and Fable 5.1 costs $10 and $50. Gemini 3.8 Flash is therefore cheaper than the cheapest frontier Claude model even after the introductory rate expires, and roughly thirteen times cheaper than Fable 5.1 on input. These are different classes of model, but for high volume summarisation and extraction the price gap is often the whole decision.

Do Claude and Gemini both work from Southeast Asia?

Yes, both, with no VPN, across Singapore, Malaysia, the Philippines, Indonesia and Thailand. Neither vendor gates access regionally and both accept Singapore-issued cards. Gemini handles Malay and Thai noticeably better than Claude does, which matters if you work in those languages rather than only in English. On data residency both store conversations in the United States, so for regulated material the question you answer first is not which model is better but whether US storage is acceptable at all.

Sources: Anthropic's help centre article on paid plan context windows for the 1M window on Fable 5.1, Opus 5 and Sonnet 5, the 500K figure on Opus 4.6 to 4.8 and Sonnet 4.6, and the Sonnet 5 compaction at 500K in Claude Cowork. Anthropic's pricing page for the free plan covering Sonnet and Haiku with Claude Code, Opus and Projects excluded, and Pro at $20 monthly or $17 annual with Max from $100. Anthropic's models overview for Sonnet 5, Opus 5 and Fable 5.1 per-token pricing, the 128K max output, batch and cache discounts, and the 555,000 word estimate for a 1M window. Google's Gemini API changelog for Gemini 3.8 Flash reaching general availability on 2 September 2026 and its positioning. Google's latest model page for the 1M context window, 64K max output, and the introductory pricing of $0.75 and $3.75 per million tokens through 31 December 2026 rising to $1.50 and $7.50. Gemini app release notes for the absence of any documented 3.8 Flash rollout to the free app tier. All figures verified 6 October 2026 and subject to change. Check each vendor directly before committing to a plan.