Google, Anthropic · The makers’ own facts

Gemini 4 Neon vs Claude Opus 5.5 (2026): price, Google’s own benchmark table, and where each is available

Gemini 4 Neon, which Google released on November 18, 2026, lists at $3 / $15 per million input / output tokens against $4 / $20 for Claude Opus 5.5, and in Google’s own model-card table Neon scores higher on most shared benchmarks while Opus 5.5 leads on Terminal-bench 4.1 and FrontierCode 2.0.

Last updated:

Use it now

Chat

The chat starts with Gemini 4 Neon; switch to Claude Opus 5.5 in the model list to ask it the same thing.

≈ 4 credits per reply · See all models

New accounts get 5 free credits once, valid for 7 days. They work on the lower-priced models here; the priciest ones need credits you bought.

Start chatting — 5 free credits

Per million tokens

Price per million tokens

checked 2026-11-19
-InputOutputOpenRouter id
Gemini 4 Neon$3$15google/gemini-4-neon
Claude Opus 5.5$4$20anthropic/claude-opus-5.5

The makers’ own facts

Side by side

-Gemini 4 NeonClaude Opus 5.5
Released
Announced November 18, 2026Source: blog.google · checked 2026-11-19
September 22, 2026; Anthropic lists retirement not sooner than September 22, 2027Source: platform.claude.com · checked 2026-11-19
Knowledge cutoff
March 2026, "though in some domains the model’s knowledge may be less up to date"Source: deepmind.google · checked 2026-11-19
Reliable knowledge cutoff June 2026 (training data cutoff also June 2026)Source: platform.claude.com · checked 2026-11-19
How the maker describes it
Frontier-level intelligence with fast end-to-end task completion for complex software engineering, knowledge work and science; sits between Gemini 4 Hydrogen and Gemini 4 Carbon for high-frequency complex tasksSource: aistudio.google.com · checked 2026-11-19
"For long-running agentic coding and knowledge work"Source: platform.claude.com · checked 2026-11-19
Price per 1M tokens (input / output)
$3.00 / $15.00 (output price includes thinking tokens)Source: aistudio.google.com · checked 2026-11-19
$4 / $20Source: platform.claude.com · checked 2026-11-19
Cached input per 1M tokens
$0.15, implicit caching only (explicit caching is not supported)Source: aistudio.google.com · checked 2026-11-19
Cache reads $0.20; cache writes $5 (5 minutes) or $8 (1 hour)Source: platform.claude.com · checked 2026-11-19
Batch
Batch and Flex: $1.50 / $7.50 per 1M tokens (input / output)Source: aistudio.google.com · checked 2026-11-19
Batch API: 50% discount on input and outputSource: platform.claude.com · checked 2026-11-19
Premium tier (Google Priority; Anthropic fast mode)
Priority: $5.40 / $27.00 per 1M tokens (input / output)Source: aistudio.google.com · checked 2026-11-19
Fast mode: $8 / $40, "up to 2.5x speed" according to AnthropicSource: anthropic.com · checked 2026-11-19
One request, 10,000 tokens in and 1,000 out (our arithmetic, list prices only)
About $0.045 (Google bills thinking tokens as output, so real requests cost more)Source: aistudio.google.com · checked 2026-11-19
About $0.06Source: platform.claude.com · checked 2026-11-19
Context window / max output
1,048,576 tokens shared by input and output; max output 65,536 by default, up to 1,048,576 with long decodeSource: aistudio.google.com · checked 2026-11-19
1M tokens / 128K tokens (300K on the Batch API with a beta header)Source: platform.claude.com · checked 2026-11-19
Input and output types
Text, image, video, audio and PDF in; text outSource: aistudio.google.com · checked 2026-11-19
Text and images in; text outSource: platform.claude.com · checked 2026-11-19
Reasoning
Four thinking levels: low, medium, high (default) and xhighSource: aistudio.google.com · checked 2026-11-19
Adaptive thinking, always on and cannot be turned off; the effort parameter controls depth, default mediumSource: platform.claude.com · checked 2026-11-19
Where you can use it
Gemini API in Google AI Studio (paid tier), Gemini for business, Google Antigravity for paid Google AI Pro and Ultra subscribers, and the Gemini app for Pro and Ultra subscribersSource: blog.google · checked 2026-11-19
Claude API, Amazon Bedrock, Claude Models on Google Cloud, Microsoft Foundry and Claude Platform on AWSSource: platform.claude.com · checked 2026-11-19
On OpenRouter
Listed as google/gemini-4-neon at $3 / $15 per million tokens (checked November 19, 2026)Source: openrouter.ai · checked 2026-11-19
Listed as anthropic/claude-opus-5.5 at $4 / $20 per million tokens (checked November 19, 2026)Source: openrouter.ai · checked 2026-11-19
Typical reply on this site (credits)
About 4 credits (a credit is about $0.01 in the $5 pack), billed by its real token cost, at least 1 creditSource: openrouter.ai · checked 2026-11-19
About 6 credits (a credit is about $0.01 in the $5 pack), billed by its real token cost, at least 1 creditSource: openrouter.ai · checked 2026-11-19
Google’s table: AA Index v5.0.1 (capability index)
61Source: deepmind.google · checked 2026-11-19
59Source: deepmind.google · checked 2026-11-19
Google’s table: Vals Index v2.2 (capability index)
70.9%Source: deepmind.google · checked 2026-11-19
70.1%Source: deepmind.google · checked 2026-11-19
Google’s table: CursorBench 4.1 (agentic coding, score)
65.4%Source: deepmind.google · checked 2026-11-19
58.1%Source: deepmind.google · checked 2026-11-19
Google’s table: Terminal-bench 4.1 (agentic coding, cheating-adjusted accuracy)
65.2%Source: deepmind.google · checked 2026-11-19
65.8%Source: deepmind.google · checked 2026-11-19
Google’s table: FrontierCode 2.0 (agentic coding, main score)
46.2%Source: deepmind.google · checked 2026-11-19
47.3%Source: deepmind.google · checked 2026-11-19
Google’s table: SWE-sweep, provider harness (agentic coding)
39.0%Source: deepmind.google · checked 2026-11-19
10.1%Source: deepmind.google · checked 2026-11-19
Google’s table: AutomationBench (knowledge work)
56.3%Source: deepmind.google · checked 2026-11-19
42.5%Source: deepmind.google · checked 2026-11-19
Google’s table: Agents’ Last Exam v1.1 (knowledge work)
75.8%Source: deepmind.google · checked 2026-11-19
71.6%Source: deepmind.google · checked 2026-11-19
Google’s table: Terminal-Bench-Science 0.1 (science and math)
70.5%Source: deepmind.google · checked 2026-11-19
63.3%Source: deepmind.google · checked 2026-11-19
Google’s table: HLE-Diamond (reasoning)
71.3%Source: deepmind.google · checked 2026-11-19
67.9%Source: deepmind.google · checked 2026-11-19
Google’s table: Chartography (multimodal)
73.5%Source: deepmind.google · checked 2026-11-19
66.3%Source: deepmind.google · checked 2026-11-19
Google’s table: GraphWalks, 256k to 1M, BFS (long context, F1)
95.2%Source: deepmind.google · checked 2026-11-19
66.8%Source: deepmind.google · checked 2026-11-19
Google’s table: WebExploitBench v1.2 (cybersecurity)
51.3%Source: deepmind.google · checked 2026-11-19
37.1%Source: deepmind.google · checked 2026-11-19
Anthropic’s launch table: Terminal-Bench-Science 0.1 (same name as above)
Not in Anthropic’s table, which was published on September 22, 2026, before Neon’s releaseSource: anthropic.com · checked 2026-11-19
58.7% (Anthropic’s own figure, at max effort)Source: anthropic.com · checked 2026-11-19
Anthropic’s launch table: AutomationBench (same name as above)
Not in Anthropic’s table (published before Neon’s release)Source: anthropic.com · checked 2026-11-19
40.0% (a Zapier run during early access, according to Anthropic’s footnote)Source: anthropic.com · checked 2026-11-19
Anthropic’s launch table: Chartography (same name as above)
Not in Anthropic’s table (published before Neon’s release)Source: anthropic.com · checked 2026-11-19
89.0%, marked "with tools" in Anthropic’s tableSource: anthropic.com · checked 2026-11-19

Which one does the maker recommend?

  • Google’s what’s-new page recommends Neon for complex software engineering, knowledge work and science where you want frontier-level intelligence with fast end-to-end task completion, Carbon for the hardest architecture, debugging, science and security problems, and Hydrogen as the default for everyday coding and agentic work.Source: aistudio.google.com · checked 2026-11-19
  • Anthropic’s models overview says to start with Claude Opus 5.5 for most workloads, and to use Claude Fable 5.2 for demanding reasoning and long-horizon agentic work, or when your own evals on Opus 5.5 at higher effort still fall short.Source: platform.claude.com · checked 2026-11-19
  • The only side-by-side of the two is on Google’s own Neon model card: Google chose the benchmarks and the rivals (Claude Opus 5.5 and GPT-6.2 Sol), and the card does not say which effort setting Opus 5.5 ran at. Neon scores higher on most rows both models have, and Opus 5.5 scores higher on Terminal-bench 4.1 (65.8% against 65.2%) and FrontierCode 2.0 (47.3% against 46.2%). Google’s figures, not tested by us.Source: deepmind.google · checked 2026-11-19
  • Anthropic’s own launch table, published before Neon existed, gives Opus 5.5 different numbers for some same-named tests: 58.7% on Terminal-Bench-Science 0.1 against 63.3% in Google’s table, and 40.0% on AutomationBench against 42.5%. Anthropic says it ran Opus 5.5 at max effort with its production safeguards on and that benchmark margins have become a less reliable guide to real-world differences. Anthropic’s Terminal-Bench, FrontierCode and CursorBench rows are different versions (4.0, v1.1, 4.0) from Google’s (4.1, 2.0, 4.1), so we do not mix them.Source: anthropic.com · checked 2026-11-19
  • Google’s launch post says Neon costs less per task than top competing frontier models, without naming Opus 5.5 in the text we read; Anthropic’s launch post compares Opus 5.5’s cost with Opus 5 and GPT-6 Astra. Neither maker’s page we read compares the cost per task of Neon and Opus 5.5 directly.Source: blog.google · checked 2026-11-19

Sample answers (AI-generated)

Same question, both models

Sample answers are pending: they have not been generated yet.

FAQ

Questions

Is Gemini Neon better than Opus 5.5?

We have not run either model, and the only side-by-side we found is Google’s own. Its model-card table puts Gemini 4 Neon ahead of Claude Opus 5.5 on most rows both have scores for, such as 70.9% against 70.1% on the Vals Index and 65.4% against 58.1% on CursorBench 4.1, but behind it on Terminal-bench 4.1 (65.2% against 65.8%) and FrontierCode 2.0 (46.2% against 47.3%). Google picked the benchmarks, and Anthropic prints different figures for some same-named tests, so read the table as Google’s case for Neon.

How much cheaper is Gemini Neon than Opus 5.5?

On list prices Neon is one quarter cheaper per token: $3 / $15 per million input / output tokens against $4 / $20. For a request with 10,000 tokens in and 1,000 out that is about $0.045 against $0.06 (our arithmetic, list prices only; Google bills thinking tokens as output, so real requests cost more). What a task really costs also depends on how many tokens each model spends, and neither maker’s page that we read compares the two on cost per task. For another Opus 5.5 price comparison, see MAI Code 2 vs Opus 5.5.

When was each released, and which knows more recent facts?

Claude Opus 5.5 came out on September 22, 2026 and Gemini 4 Neon on November 18, 2026, 57 days later (our arithmetic). Google gives Neon a knowledge cutoff of March 2026, with the warning that some domains may be less up to date; Anthropic lists June 2026 as Opus 5.5’s reliable knowledge cutoff. Anthropic says Opus 5.5 will not be retired before September 22, 2027.

Where can I use each, and can I chat with both here?

Google lists Neon in the Gemini API in Google AI Studio (paid tier), Gemini for business, Google Antigravity for paid AI Pro and Ultra plans, and the Gemini app for Pro and Ultra subscribers; its Google Cloud page says verified enterprise customers in select regions. Anthropic lists Opus 5.5 on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS. Both are on OpenRouter and both run here: a typical reply to Gemini 4 Neon costs about 4 credits and a typical reply to Claude Opus 5.5 about 6 credits (a credit is about $0.01 in the $5 pack), billed by its real token cost, at least 1 credit.

Which should I pick, Gemini Neon or Opus 5.5?

It depends on cost and on what each maker says the model is for. Neon is $3 / $15 and Google positions it between Hydrogen and Carbon for complex, high-frequency software engineering, knowledge work and science; Opus 5.5 is $4 / $20, and Anthropic positions it for long-running agentic coding and knowledge work and says to start with it for most workloads. If list price per token matters most, Neon is cheaper; for long agentic coding, look at the Terminal-bench 4.1 and FrontierCode 2.0 rows, where Google’s own card has Opus 5.5 slightly ahead. The surest test is your own task on both here. Google’s other models are on Gemini 4 Hydrogen and Gemini 4 Carbon.

Which Opus does “Gemini Neon vs Opus” mean?

Google’s model card compares Neon with Claude Opus 5.5, and Anthropic’s model page marks Opus 5.5 as the latest Opus, so this page uses that one. Anthropic’s next model up is Claude Fable 5.2 at $8 / $40; see Claude Fable 5.2 vs Opus 5.5. Google’s whole line is on Google Gemini.

Is this an official Google or Anthropic page?

No. This page is unofficial and not affiliated with or endorsed by Google or Anthropic. Every fact links to the page we read.