Gemini 4 Carbon vs Claude Sonnet 5.5
Gemini 4 Carbon scores higher on the Models at Work leaderboard (77 vs 72 out of 100). On independent tests alone, Gemini 4 Carbon is ahead (97 vs 83). They cost about the same: 1,000 support replies come to $10 on Gemini 4 Carbon and $10 on Claude Sonnet 5.5 at list price. On the Artificial Analysis Intelligence Index, one of the independent tests, Gemini 4 Carbon leads 60 to 55.
| Gemini 4 Carbon · Google | Claude Sonnet 5.5 · Anthropic | |
|---|---|---|
| Leaderboard rank | #3 | #6 |
| Score (0–100) | 77 | 72 |
| Kind of work | Hardest problems | Everyday default |
| Price per 1M tokens | $2 to read · $10 to write | $2 to read · $10 to write |
| Reads at once | 1M tokens (about 750,000 words) | 1M tokens (about 750,000 words) |
| Where you can get it | Gemini Enterprise Agent Platform (Google Cloud), Google AI Studio, Gemini for business, Gemini app (Google AI Ultra) | Claude Team and Enterprise, Amazon Bedrock, Gemini Enterprise Agent Platform (Google Cloud), Microsoft Foundry, Microsoft 365 Copilot, GitHub Copilot |
| Can a company host it itself | No | No |
| Artificial Analysis Intelligence Index | 60 | 55 |
| Vals Finance Index | 67.05% (rank 1) | 64.53% (rank 6) |
| Vals Tax Agent Bench | 51.78% (rank 2) | 41.96% (rank 12) |
| Vals Legal Research Bench | 55.77% (rank 1) | 48.08% (rank 15) |
| Professional Reasoning Benchmark - Legal (Scale) | 53.35 | Not measured |
| Humanity's Sixth Sense (Scale) | 71.10 | Not measured |
| Humanity's Last Exam, Diamond (Scale) | Not measured | 37.50 |
| Best for | Reading and comparing large sets of dense documents, Legal and finance analysis, Multi-step research | Support replies at volume, Drafting proposals and reports, Summarising meetings and long documents, Answering questions from company documents |
Cost of 1,000 jobs
| 1,000 of these | Gemini 4 Carbon | Claude Sonnet 5.5 |
|---|---|---|
| Support repliesthe customer message, the account notes and two help articles read; a 300-word reply written | $10 | $10 |
| Documents processeda two-page invoice or form read; the fields pulled out | $7 | $7 |
| Calls summariseda 40-minute call transcript read; a summary and next steps written | $26 | $26 |
| Contracts revieweda 40-page contract read against a checklist; a memo of issues written | $75 | $75 |
| Drafts writtena brief and one example read; a 600-word email or proposal section written | $11 | $11 |
Worked out from each vendor’s list price and a fixed size for each job. A product built on a model charges its own price on top. See what every model costs for the same jobs.
Choose Gemini 4 Carbon for
- Reading and comparing large sets of dense documents
- Legal and finance analysis
- Multi-step research
First on Vals AI's tests of finance work (67.1%) and legal research (55.8%) and second on Artificial Analysis's index (60 against the leader's 65), at a quarter of the leader's list price until 30 November, when Google doubles it.
Caveats: The $2 in / $10 out per million tokens (the unit vendors bill by) is introductory; Google says $4 / $20 applies after 30 November 2026, which will lower its score here. Google says it uses more text per task than Gemini 4 Argon, so the bill can run higher at the same list price. The rankings quoted in Google's launch post are relayed by Google and are not counted; six independent tests are. In the Gemini app it is for Google AI Ultra subscribers only, per Google; commenters on lower plans and one enterprise account report they cannot select it yet. It ranks 3rd, behind two cheaper models that are within three points of it: a tie. Expect it to fall when the price doubles on 1 December.
Choose Claude Sonnet 5.5 for
- Support replies at volume
- Drafting proposals and reports
- Summarising meetings and long documents
- Answering questions from company documents
A sound everyday default: within a point of Claude Opus 5.5 on Vals AI's test of finance work (64.5% against 65.3%) at half its list price.
Caveats: The saving depends on the effort setting: at the top setting Artificial Analysis's test run cost $10.02 per task, level with Opus 5.5 ($10.04). On Scale's Humanity's Last Exam, a set of expert-level questions, it scores 37.5 against 55.0 for Opus 5.5: move up for the hardest analysis. In Microsoft 365 Copilot it has been rolling out since 30 September with unpublished usage limits; Claude models are off in the EU, EFTA and UK until an administrator allows Anthropic as a subprocessor (a firm allowed to handle your data). Vals AI notes that some of its test tasks were answered by a fallback model after Sonnet 5.5 refused them. Up from 11th to 6th because Vals AI's finance, tax and legal research tests now count toward the score. Places 6 to 8 share a score.
What people running them say
Gemini 4 Carbon: On the Hacker News (a developer forum) thread about Google's launch, several commenters on the Pro and Plus plans say Carbon shows in the app as Ultra-only, one says their company's enterprise account cannot use it yet, and another says their workplace already finds Argon too expensive. None of this is verified, and it is too little to score.
Claude Sonnet 5.5: Less discussed than Opus or Haiku because it just works; the news was October's price cut for text it has already been sent once and reads again, to $0.10 per million tokens (the unit vendors bill by).
Questions people ask
Which is better, Gemini 4 Carbon or Claude Sonnet 5.5?
Gemini 4 Carbon scores higher on the Models at Work leaderboard (77 vs 72 out of 100). On independent tests alone, Gemini 4 Carbon is ahead (97 vs 83). They cost about the same: 1,000 support replies come to $10 on Gemini 4 Carbon and $10 on Claude Sonnet 5.5 at list price. On the Artificial Analysis Intelligence Index, one of the independent tests, Gemini 4 Carbon leads 60 to 55.
When should I choose Gemini 4 Carbon over Claude Sonnet 5.5?
Choose Gemini 4 Carbon for reading and comparing large sets of dense documents, legal and finance analysis, multi-step research. First on Vals AI's tests of finance work (67.1%) and legal research (55.8%) and second on Artificial Analysis's index (60 against the leader's 65), at a quarter of the leader's list price until 30 November, when Google doubles it.
When should I choose Claude Sonnet 5.5 over Gemini 4 Carbon?
Choose Claude Sonnet 5.5 for support replies at volume, drafting proposals and reports, summarising meetings and long documents, answering questions from company documents. A sound everyday default: within a point of Claude Opus 5.5 on Vals AI's test of finance work (64.5% against 65.3%) at half its list price.
Which is cheaper, Gemini 4 Carbon or Claude Sonnet 5.5?
Gemini 4 Carbon costs $2 per million tokens it reads (input) and $10 per million tokens it writes (output) on Google's list price (Google: Gemini 4 Carbon is now available, read Nov 18, 2026). A token is about three-quarters of a word. Claude Sonnet 5.5 costs $2 per million tokens it reads (input) and $10 per million tokens it writes (output) on Anthropic's list price (Claude pricing, read Nov 18, 2026). A token is about three-quarters of a word.
What do 1,000 jobs cost on Gemini 4 Carbon and Claude Sonnet 5.5?
At list price: 1,000 support replies cost $10 on Gemini 4 Carbon and $10 on Claude Sonnet 5.5; 1,000 documents processed cost $7 on Gemini 4 Carbon and $7 on Claude Sonnet 5.5; 1,000 calls summarised cost $26 on Gemini 4 Carbon and $26 on Claude Sonnet 5.5; 1,000 contracts reviewed cost $75 on Gemini 4 Carbon and $75 on Claude Sonnet 5.5; 1,000 drafts written cost $11 on Gemini 4 Carbon and $11 on Claude Sonnet 5.5. A product built on either model charges its own price on top.
Scores come from the Models at Work leaderboard: Score is 0 to 100 and computed, not typed. 40% is independent testing (Artificial Analysis, Arena, Scale, Epoch, Vals and others; each test is scaled against the best score on this board, and tests of real work, such as finished freelance projects or professional tasks, count double the exams). 30% is price: the vendor's list price for a typical mix of reading and writing, on a fixed scale from $0.05 to $60 per million tokens, so a model that costs a fifth as much gains real ground. 15% is what people report after running it, from Signals and linked threads. 15% is fit: how much text it can take in at once and where a business can get it (the products and clouds that carry it), with a small credit for open weights. A missing component drops out and the rest are reweighted. A model with no independent test yet is provisional and ranks below every measured one. Within three points is a tie. Vendor-published figures are labelled and never counted. Prices are vendor list prices where published; a typical mix assumes three parts reading to one part writing.