Claude Sonnet 5: long context and agentic work

Claude Sonnet 5 is an Anthropic model released on June 30, 2026, and the most agentic Sonnet in the line. On Genosai it handles up to 800,000 tokens of context, nearly seven times more than the previous Sonnet 4.6. Billing is by tokens: 190 ₽ per 1M input and 940 ₽ per 1M output, which is cheaper than the older version.

Updated: July 23, 2026

• 800,000-token context — The largest window among Claude models on Genosai: a book, a codebase, or a year of correspondence at once.
• Cheaper than the last Sonnet — Client price 190/940 ₽ per 1M versus 280/1400 ₽ for Sonnet 4.6, so the newer generation actually costs less.
• Agentic work — Plans steps, calls tools, and drives browsers and terminals across long multi-step scenarios.
• Vision and files — Accepts images and attachments in JPG, PNG, and WebP formats, up to 10 files per request.
• Opus-level quality — Anthropic places it close to the Opus 4.8 flagship while keeping the speed of a mid-tier model.

Contents

• What is Claude Sonnet 5
• Capabilities
• Examples prompt and response
• How to use on Genosai
• Prompts
• Generation cost
• How it compares
• Limitations and tips
• FAQ

What is Claude Sonnet 5

Claude Sonnet 5 is an Anthropic text model released on June 30, 2026. It opens the fifth generation of the Sonnet line and is described by its developer as the most agentic Sonnet to date: the model makes plans, uses tools such as browsers and terminals, and completes long tasks autonomously. Only a few months ago that level of independence required larger and more expensive models.

The main difference from the previous Sonnet 4.6 is not a single feature but a shift across every key direction at once. Anthropic calls the improvement substantial in reasoning, tool use, coding, and knowledge work. It separately reports that the model strictly outperforms Sonnet 4.6 in agentic search and computer use. The positioning speaks for itself too: in the Claude apps it became the default model on the Free and Pro plans.

On Genosai, Sonnet 5 has something no other Claude model on the platform offers, an 800,000-token context window. Sonnet 4.6 and the Opus flagships work with 120,000 tokens, so the new model holds nearly seven times more material in a single pass. That changes the usage pattern: instead of slicing a document into chunks and stitching the conclusions together, you hand it over whole.

The second surprise is price. A new generation usually costs more, but here it is the opposite: Sonnet 5 costs 190/940 ₽ per 1M tokens versus 280/1400 ₽ for Sonnet 4.6. The result is a rare combination — newer, smarter, with a window nearly seven times wider, and cheaper than its predecessor. That is why it becomes a sensible starting point for most everyday tasks rather than a compromise between quality and budget.

Capabilities

Claude Sonnet 5 is equally confident with text, code, data, and images.

Long context and large materials

An 800,000-token window is on the order of several hundred pages of text per request. In practice that means an annual report, a body of regulations, a set of interview transcripts, or a large code project fits whole, with no pre-slicing. The model sees all the material at once, so it can cross-reference facts from the beginning and the end of a document, find contradictions between sections, and build a summary that rests on the full volume rather than a randomly chosen fragment. For analysis and legal work this is a fundamental difference: output quality stops depending on how well you happened to cut the source.

Agentic scenarios and tool use

Anthropic puts the main emphasis on agentic behavior. Sonnet 5 builds a multi-step plan, calls tools, and carries the chain to the end instead of losing the thread on the third action. Agentic search and work with computer interfaces are both improved. For a user this shows up as a simple effect: a complex task can be stated in one request and come back finished, rather than requiring you to walk the model through every stage. It is most noticeable on tasks of the form "work through this material, compare the options, and propose a solution with justification."

Code and development

The model writes, reviews, and debugs code, explains its edits, and walks through edge cases. Combined with the large window, this lets you work with a project rather than a single file: show the whole structure and ask where the logic breaks or where the same mistake repeats. Anthropic names coding as one of the four areas where Sonnet 5 substantially outperforms the previous version.

Vision and files

Sonnet 5 accepts images in JPG, PNG, and WebP formats, up to 10 files per request. You can attach an interface screenshot and ask it to describe the problem, hand over a diagram for analysis, or pull text from a photo of a document. This pairs well with a text task: attach a mockup and ask for copy to match it, or check whether the content of an image agrees with the text next to it.

Examples prompt and response

Below are real Claude Sonnet 5 responses produced on Genosai. The first is marketing copy, the second is a code review with an edge-case breakdown. Texts are quoted verbatim, with the long answer shortened.

Prompt: Write a product-card description for a marketplace: wireless noise-cancelling earbuds, up to 30 hours battery. 3–4 sentences, no clichés.

Claude Sonnet 5 response:
**Wireless Noise-Cancelling Earbuds — 30-Hour Battery Life**

Active noise cancellation cuts ambient sound so calls and music stay clear in traffic, offices, or on flights. A single charge delivers up to 8 hours of playback, extended to 30 hours with the charging case. Touch controls handle calls, volume, and ANC toggling without reaching for your phone. IPX5 water resistance covers rain and workouts.

_Generated on Genosai.io with Claude Sonnet 5._

Prompt: Review this Python function: def avg(x): return sum(x)/len(x) — what edge cases does it miss and how to fix it?

Claude Sonnet 5 response:
# Review of `avg(x)`

## Missing Edge Cases

1. **Empty sequence** — `len(x) == 0` raises `ZeroDivisionError`
2. **Non-numeric elements** — mixed types (e.g., strings in the list) raise `TypeError` inside `sum()`
3. **`None` values** — will raise `TypeError` when summed with numbers
4. **Non-iterable / no `__len__` input** — e.g., a generator has no `len()`, raising `TypeError`
5. **NaN values** — silently produce `NaN` as output (not an error, but often unexpected)
6. **Very large inputs** — no issue with floats generally, but for exact fractions with `Decimal`/`Fraction` types, mixing with plain floats could cause precision issues

_Generated on Genosai.io with Claude Sonnet 5._

How to use on Genosai

1. Open the model catalog on Genosai and pick Claude Sonnet 5 among the text models.
2. Describe the task in the chat: goal, tone, format, and answer length.
3. Attach files if needed — a document, data, or an image — and refer to them in the request.
4. Send the request and read the answer as it streams, without waiting for the full generation.
5. Refine in the same thread: ask to shorten, add an example, or restructure.

The large context window changes the usual workflow. Long material used to be cut into parts, processed piece by piece, and manually stitched back together, losing the links between sections. With Sonnet 5 the logic is different: hand over the whole source and ask your question the way you would ask a colleague who has read the entire document. Genosai runs in the browser with no installation, signing up takes a minute, and billing is by actual usage, so you can start with a single request and judge the model on your own task before wiring it into a routine process. Chat history is saved, so a long analysis is easy to continue later in the same thread.

Prompts

The prompts below play to Sonnet 5's strengths: large inputs, agentic logic, and careful code. The general principle is to specify not only the topic but the structure of the answer: sections, number of points, table format. The model holds those constraints well and returns a result that is ready to use without rework.

Read the attached document in full and build a 10-point summary. Separately list any contradictions between sections.

Draft a 5-step plan for solving this task; for each step state what data is required and what result is expected.

Refactor this function: add edge-case handling and type hints. Explain each change on its own line.

Compare the two attached documents and build a table of differences. Columns: section, before, after, significance.

Based on all the text above, write an FAQ of 7 questions a real reader would ask. Answers of 2–3 sentences.

Analyze the attached interface screenshot and describe what is happening and which actions are available to the user.

Check the text for factual errors and flag the places that need a source check.

Generation cost

On Genosai, Claude Sonnet 5 is billed by tokens — you pay for the actual size of the request and the response, so a short question costs less than a long analysis. The client price is 190 ₽ per 1M input tokens and 940 ₽ per 1M output tokens. The exact amount depends on prompt length and answer volume, and it naturally grows on large documents: the 800,000-token window opens those scenarios, but it is billed by real usage.

The practical takeaway is simple. If your task fits in a couple of paragraphs, you are not overpaying for the long context, because billing counts exactly what was sent. And if you genuinely load a multi-page document, compare the cost with the alternative: splitting the material, making several requests, and merging the results by hand usually costs more and takes longer.

A starter balance after sign-up lets you try Claude Sonnet 5 for free. Current rates and your balance are in the Pricing section.

How it compares

Claude Sonnet 5 sits in an unusual spot in the line: mid-tier on price, and the leader on context among Claude models on the platform.

The previous generation, Claude Sonnet 4.6, now loses to the new version on every measurable platform parameter including price, so there is little reason to hesitate about switching. For the most critical agentic tasks and complex work in large codebases, step up to Claude Opus 4.8. If your tasks are simple and speed and economy matter more, step down to Claude Haiku 4.5. Among same-class rivals, compare GPT-5.5.

A practical rule of thumb: start with Sonnet 5 and switch only where you feel a limit. On Genosai all models are available in one account, so switching takes seconds and does not require separate subscriptions per model — you can run the same request through two models and compare the answers directly.

Limitations and tips

Anthropic points to a clear weak spot: on cybersecurity tasks Sonnet 5 is much weaker than the Opus models. If your work involves vulnerability analysis and demands maximum depth, use the flagship.

A large context window is no excuse to ignore input quality. The model sees all the material, but if it is full of noise, duplicates, and outdated versions, the answer inherits that. Before uploading, strip the obviously irrelevant parts and state explicitly which sections take priority.

A practical tip: phrase a long-document task as a question rather than a "summarize this" command. Questions like "where do the payment terms in this contract disagree" or "which requirements repeat across sections" produce far more useful results than a general request to retell.

One more technique is to state quality criteria directly in the request. Describe which phrasings to avoid, what counts as a good answer, and which data must be preserved. Sonnet 5 follows such instructions well, so the more precisely you describe the expected result, the fewer iterations you spend on rework. And if the topic needs fresh or exact data, ask the model to flag the places that need a source check.

FAQ

What is Claude Sonnet 5?

It is an Anthropic text model released on June 30, 2026. Anthropic positions it as the most agentic Sonnet yet: it plans tasks, uses tools, and runs autonomously at a level that used to require larger models. In the Claude apps it became the default model on the Free and Pro plans.

What is the context window on Genosai?

Up to 800,000 tokens per request, the largest window among Claude models on the platform. For comparison, Sonnet 4.6 and Opus 4.8 work with 120,000 tokens. That is enough for several hundred pages of text or an entire code project in one pass.

How is Claude Sonnet 5 better than Sonnet 4.6?

Anthropic calls the improvement substantial across four areas at once: reasoning, tool use, coding, and knowledge work. It also reports a strict improvement in agentic search and computer use. On Genosai there is a practical bonus on top: nearly seven times more context at a lower price.

How much does a Claude Sonnet 5 request cost?

Billing is by tokens: for the client, 190 ₽ per 1M input tokens and 940 ₽ per 1M output, with the exact amount depending on prompt and response length. That is cheaper than Sonnet 4.6 at 280/1400 ₽ and more than twice as affordable as the Opus 4.8 flagship.

Does Claude Sonnet 5 work with images?

Yes, the model accepts images in JPG, PNG, and WebP formats, up to 10 files per request. You can ask it to analyze an interface screenshot, read a diagram, or pull text from a photo of a document.

When should you pick Opus 4.8 instead?

Choose Opus 4.8 for the most critical agentic scenarios and complex work in large codebases where the cost of a mistake is high. Anthropic also notes that Sonnet 5 is much weaker than the Opus models on cybersecurity tasks. Everywhere else Sonnet 5 delivers close quality for less money and with a far larger context.

Try Claude Sonnet 5 on Genosai