# Claude Sonnet 5 vs Gemini 3.5 Pro (July 2026)

> Claude Sonnet 5 and Gemini 3.5 Pro are the two models developers cite most for accurate docs. Sonnet 5 is GA at $2/$10 with a 1M-token window; Gemini 3.5 Pro targets 2M tokens but is still in preview. Here's which to pick, scored on Falconer's Docs-bench.

- Date: 2026-07-10
- Tags: ai-models, documentation, benchmarks

---
Claude Sonnet 5 and Gemini 3.5 Pro are the two models developers cite most for accurate documentation and code review, which is why they are the decision pair worth studying if you care about docs quality. Sonnet 5 is generally available, priced at $2 input / $10 output per million tokens through August, and ships with a 1M-token context window. Gemini 3.5 Pro targets a 2M-token context window and Deep Think reasoning but remains in limited preview as of July 7 with pricing unpublished. If you need to generate and maintain docs today on a reproducible model, Sonnet 5 is the practical pick. If your codebase is large enough that context window is the binding constraint, Gemini 3.5 Pro is worth testing once it reaches general availability.

### TLDR

- Sonnet 5 (Anthropic, June 30) is generally available; Gemini 3.5 Pro (Google) is still in limited preview as of July 7.
- Sonnet 5 costs $2 input / $10 output per 1M tokens through August 31, then $3 / $15; Gemini 3.5 Pro pricing is unpublished and estimated near $15 / $60.
- Gemini 3.5 Pro targets a 2M-token context window against Sonnet 5's 1M, which matters most on large monorepos.
- Sonnet 5 uses a new tokenizer that produces about 30% more tokens for the same text (up to roughly 1.4x for English), so re-profile prompts when comparing cost.
- Falconer's Docs-bench runs both models through Spark on one codebase and scores the output with Knowledge Health, so the result reflects docs accuracy rather than reputation.

## What is Claude Sonnet 5?

Claude Sonnet 5 is Anthropic's mid-tier agentic model, [released June 30, 2026 to replace Sonnet 4.6](https://techcrunch.com/2026/06/30/anthropic-launches-claude-sonnet-5-as-a-cheaper-way-to-run-agents/). It ships with a [1M-token context window](https://platform.claude.com/docs/en/about-claude/models/whats-new-sonnet-5) and launched at promotional pricing of $2 input / $10 output per million tokens through August 31, moving to $3 / $15 after that ([Anthropic pricing](https://platform.claude.com/docs/en/about-claude/pricing)). Anthropic positions it as a [cheaper way to run agents](https://techcrunch.com/2026/06/30/anthropic-launches-claude-sonnet-5-as-a-cheaper-way-to-run-agents/), and it is the Anthropic model most developers reach for on everyday coding and documentation work.

## What is Gemini 3.5 Pro?

Gemini 3.5 Pro is Google's frontier model, announced at Google I/O on May 19, 2026, [targeting a 2M-token context window, Deep Think reasoning, and frontier multimodal capability](https://www.techtimes.com/articles/317919/20260606/google-gemini-35-pro-nears-june-launch-2-million-token-context-deep-think-reasoning.htm). As of July 7 it is still in limited enterprise preview rather than general availability, and Google has not published Pro pricing, so any cost figure is an estimate. Google is now [targeting July 17 for general availability](https://www.techtimes.com/articles/319877/20260708/gemini-35-pro-targets-july-17-deepseeks-july-24-deadline-hits-developers-now.htm), after the date slipped from June. Its headline advantage for documentation is the large context window, which lets it hold more of a repository in a single pass.

![](https://falconer.com/api/file/s3/images/1783701745220-m5shj.png)

## Feature comparison

| Dimension | Claude Sonnet 5 | Gemini 3.5 Pro |
| --- | --- | --- |
| Lab | Anthropic | Google |
| Released | June 30, 2026 | Announced May 19; preview |
| Generally available | Yes | No, limited preview |
| Input $/1M | $2 intro, then $3 | \~$15 (estimated) |
| Output $/1M | $10 intro, then $15 | \~$60 (estimated) |
| Context window | 1M tokens | 2M tokens (target) |
| Published pricing | Yes | No |
| Reproducible benchmark | Yes | Provisional until GA |

## Which one is more accurate on documentation?

Both models earned their reputation on documentation and code review, so the gap is narrow enough that a general benchmark will not settle it. The dimension that matters is accuracy against your own code, which is what Knowledge Health weights most, because a confidently wrong docs set costs a team more than a thin one. The only reliable read is to run both through Docs-bench on your repository and compare the composite scores directly.

## Which one has the bigger context window, and does it matter?

Gemini 3.5 Pro targets 2M tokens against Sonnet 5's 1M, so on a large monorepo it can take in roughly twice the source in a single pass. For a from-scratch docs run this can raise coverage, since the model sees more of the surface area at once and misses fewer modules. On a service-sized repository that fits comfortably inside 1M tokens, the window difference stops mattering and the decision comes down to accuracy and cost.

## Which one is cheaper to run?

Sonnet 5 has published, competitive pricing at $2 input / $10 output through August. Gemini 3.5 Pro has no published Pro pricing, and the circulating estimate near $15 / $60 would make it several times more expensive per docs run. Two caveats cut against Sonnet 5's headline: its intro rate expires August 31, and its [new tokenizer produces about 30% more tokens for the same text](https://simonwillison.net/2026/Jun/30/claude-sonnet-5/), up to roughly 1.4x for English. Even with both caveats, Sonnet 5 is the cheaper and more predictable option today. Cost only matters relative to output quality, and the lever most teams underuse is context: feed either model an accurate, current view of the codebase and the cheaper option usually clears the bar. That is where [Falconer](https://falconer.com/) helps, keeping docs grounded in your current code so a cheaper model has better context to work from.

### Which one can you build a reproducible workflow on?

Sonnet 5 is generally available, so a documentation pipeline built on it is stable and a Docs-bench run against it can be repeated. Gemini 3.5 Pro is preview-gated, its pricing is unpublished, and its general availability date has already slipped from June to July. A preview model can change between runs, so any documentation workflow that depends on Gemini 3.5 Pro today carries reproducibility risk until Google ships general availability.

### When Claude Sonnet 5 is the right choice

Pick Sonnet 5 if you need to generate and maintain docs now, want published and predictable pricing, and value a model you can build a repeatable pipeline on. For most engineering teams shipping a docs set this quarter, it is the safer default.

### When Gemini 3.5 Pro is the right choice

Pick Gemini 3.5 Pro if your repository is large enough that a 1M-token window forces you to chunk the codebase, and you can wait for general availability or already have enterprise preview access. The 2M-token window is the reason to test it, and coverage on very large monorepos is where it can pull ahead.

## How to choose between Sonnet 5 and Gemini 3.5 Pro

Decide on two questions. First, does your codebase exceed a 1M-token window in the docs task you care about? If not, the window advantage is moot and Sonnet 5 wins on availability and price. Second, do you need to ship a reproducible workflow now? If yes, Sonnet 5 is the only one of the two that is generally available today. Run both through Docs-bench on your own repo before committing, since the accuracy gap is narrow and codebase-specific.

![](https://falconer.com/api/file/s3/images/1783701771212-txjufs.png)

Want to test both on your codebase with a fixed scoring config? Read the [Docs-bench methodology](https://falconer.com/benchmarks?utm_source=blog&utm_medium=organic&utm_campaign=sonnet5-vs-gemini35pro).

## See both models scored on your repo

The model you pick sets a ceiling on docs quality. What you feed it decides whether you hit that ceiling. Falconer runs Docs-bench on the model you choose, then keeps the docs current as your code changes, turning them into [a company brain](https://falconer.com/guides/what-is-company-brain/) that reflects your repository rather than a leaderboard.

- **Grounded in your code.** Answers and generated docs pull from your actual codebase, Slack, and tickets, not the model's general training.

- **Docs that update themselves.** When a PR merges, Falconer reads the diff, finds the affected docs, and proposes the edits, then pings the owner in Slack to accept or revise.

- **Knowledge Health scoring.** The same Coherence, Coverage, Freshness, and Density checks behind Docs-bench run on your own knowledge base, so you see what is stale, missing, or contradictory.

- **Query it anywhere.** Reach the same current source from the editor, Slack, and coding agents like Claude Code and Cursor over MCP.

[See the latest Docs-bench results](https://falconer.com/benchmarks?utm_source=blog&utm_medium=organic&utm_campaign=sonnet5-vs-gemini35pro), or read the [public head-to-head writeup](https://huggingface.co/blog/maxifalconer/falconer-notion-confluence-benchmarks) scoring Falconer against Notion, Atlassian Rovo, Claude Code, and Codex on 200 real questions.

## FAQ

### Is Claude Sonnet 5 or Gemini 3.5 Pro better for documentation? 

Both are the models developers cite most for documentation accuracy, so the gap is narrow. Sonnet 5 is the practical pick today because it is generally available and priced at $2 input per million tokens through August, while Gemini 3.5 Pro is still in limited preview. The larger lever is context: Falconer feeds either model your current code and decisions, which moves accuracy more than the model choice does.

### Is Gemini 3.5 Pro available yet? 

As of July 7, 2026, Gemini 3.5 Pro is in limited enterprise preview, not general availability. Google announced it at I/O on May 19, the general availability target slipped from June, and Google is now targeting July 17. If you build on Falconer, the model is a setting, so you can switch to Gemini 3.5 Pro the day it ships GA without rebuilding your pipeline.

### How much does Gemini 3.5 Pro cost? 

Google has not published Pro pricing. Circulating estimates put it near $15 input / $60 output per million tokens, but treat any figure as provisional until general availability. Whatever the sticker, the cheapest run is the one that gets the answer right the first time, which is why Falconer grounds each run in your current code instead of paying a model to rediscover it.

### Does the bigger context window make Gemini 3.5 Pro better for docs? 

Only when your codebase does not fit inside Sonnet 5's 1M-token window. The 2M-token target helps coverage on large monorepos; on smaller repositories the difference disappears and accuracy and cost decide. A bigger window also matters less when Falconer retrieves only the code a task touches, so the model rarely needs the whole repo in one pass.

### Which model should I build a documentation pipeline on today? 

Claude Sonnet 5, because it is generally available with published pricing and can support a reproducible workflow. Gemini 3.5 Pro carries reproducibility risk until it exits preview. Falconer runs on Sonnet 5 today and lets you swap the model later, so you get a stable pipeline now without locking out Gemini once it ships.

### Does Sonnet 5's new tokenizer change what I actually pay? 

Yes. The per-token price is competitive, but the new tokenizer produces about 30% more tokens for the same text, up to roughly 1.4x for English, so run your own prompts through it before trusting a headline cost comparison. Tighter context keeps that tax small, which is one reason Falconer sends the model only the code a task needs rather than the whole file tree.

### How do I keep generated docs accurate after the model writes them?

Generation is the easy half. Docs drift the moment the next PR merges. Falconer closes that loop by reading each merged diff, finding the docs it affects, and proposing the edit, so accuracy does not depend on anyone remembering to update a page. That is the core of [a self-updating company brain](https://falconer.com/guides/self-updating-company-brain/).