Updated October 28, 2026. Rank says Gemini 4 Carbon is ahead. Price, access, and your workload can each overturn that. Public scores include evidence status and uncertainty.
Gemini 4 Carbon has the higher public point estimate, 88.42 versus 81.86. Their conditional score ranges overlap. These ranges do not establish rank confidence. 3 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.
Recommendations appear only when a shared evidence basis or an explicit operating constraint supports the call. Secondary and unsupported use cases stay disclosed below the initial list.
Code generation, repair, and software-engineering tasks
Claude Sonnet 5.5
Claude Sonnet 5.5 has the higher public coding point estimate, 83.8 to 82.2, with Supported evidence for both models. The conditional ranges do not establish rank confidence.
Tool use, computer use, and multi-step task completion
Not enough matched evidence
Gemini 4 Carbon is not ranked on the public lane for agentic, so no winner is named for agentic.
Prompts that approach the documented context limit
Not enough matched evidence
A complete context comparison is not sourced.
1K fresh input + 500 output tokens
No clear pick
A complete comparable API-rate estimate is not available for both models.
200K cached + 20K fresh input + 10K output tokens
No clear pick
A complete comparable API-rate estimate is not available for both models.
50K fresh input + 3K output tokens
No clear pick
A complete comparable API-rate estimate is not available for both models.
Choose a job from the LLM Selector's task catalog to see the category row it rests on, under the same basis rules as the table below. A directional row stays directional; choosing a job never creates a winner.
The same task catalog as the LLM Selector. Each job names the evidence surface it rests on; nothing here adds a new score.
Like-for-like · BenchAlign v5.8
Claude Sonnet 5.5 has the higher coding point estimate. Conditional score ranges do not establish rank confidence.
Coding scores combine specific tasks and setups. Match the editor, harness, and effort to your workflow.
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
Only shared public evidence is shown. Sparse evidence stays a ruled list rather than being closed into a radar shape.
A shared-evidence shape is not available.
BenchLM does not draw a radar or infer missing axes when the matched evidence is too sparse.
Each row shows the public-lane category score for both models: the BenchAlign v5.8 lane for agentic, coding, and knowledge, and the weighted public rows elsewhere, with the same rank each model page reports. A row is like-for-like only when both scores rest on Supported evidence or the same weighted set. Directional and not-comparable rows remain visible, but they never receive a winner.
| Category | Claude Sonnet 5.5 | Gemini 4 Carbon | Basis | Reading |
|---|---|---|---|---|
| Coding | 83.8Supported · #1/156 | 82.2Supported · #3/156 | Like-for-likeBenchAlign v5.8 lane · 11 vs 1 public rows | Claude Sonnet 5.5 leads · intervals overlap |
| Knowledge | 81.6Supported · #7/178 | 88.2Supported · #2/178 | Like-for-likeBenchAlign v5.8 lane · 16 vs 0 public rows | Gemini 4 Carbon leads · intervals overlap |
| Agentic | 69.0Supported · #11/134 | Not ranked | Not comparableBenchAlign v5.8 lane · 16 vs 4 public rows | Not comparable |
| Reasoning | 81.8#6/26 | 79.5Unranked · 2 rankable rows | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| Multimodal | 85.1Unranked · 7 rankable rows | Not ranked | Not comparableProvisional lane · 1 vs 0 weighted rows | Not comparable |
| Multilingual | Not ranked | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| Instruction following | Not ranked | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| Math | 95.2Unranked · 4 rankable rows | Not ranked | Not comparableProvisional lane · 2 vs 0 weighted rows | Not comparable |
Ranks count the models scored in each category’s lane, so the agentic, coding, and knowledge denominators (BenchAlign v5.8) differ from the provisional-lane categories. Unranked scores sit on the provisional lane’s scale but never name a winner. Like-for-like rows within 0.5 points read as a practical tie.
Three fixed token mixes turn per-token rates into comparable decisions. Each scenario states context fit and whether cached input had to fall back to the published list-input rate.
1K fresh input + 500 output tokens
Modeled costs are equal
Costs use the listed standard API rates.
50K fresh input + 3K output tokens
Modeled costs are equal
Costs use the listed standard API rates.
200K cached + 20K fresh input + 10K output tokens
Modeled costs are equal
Costs use the listed standard API rates.
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
Maximum documented context; output-token limits may be lower.
Claude Sonnet 5.5
Gemini 4 Carbon
Not sourced
Google Gemini 4 Carbon launchClaude Sonnet 5.5
claude-sonnet-5-5
Anthropic Claude Sonnet 5.5 model documentationGemini 4 Carbon
Not sourced
A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
Claude Sonnet 5.5
$0.1 per 1M cached input tokens
Claude API pricingGemini 4 Carbon
$0.1 per 1M cached input tokens
Google Gemini 4 Carbon launch postClaude Sonnet 5.5
Gemini 4 Carbon
Not sourced
Claude Sonnet 5.5
Gemini 4 Carbon
Not sourced
Claude Sonnet 5.5
Generally Available · Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS
Anthropic Claude Sonnet 5.5 model documentationGemini 4 Carbon
Limited Access · Fairwind Program
Google Gemini 4 Carbon launchClaude Sonnet 5.5
Reasoning
Gemini 4 Carbon
Reasoning
Claude Sonnet 5.5
Proprietary
Gemini 4 Carbon
Proprietary
Claude Sonnet 5.5
Proprietary
Gemini 4 Carbon
Proprietary
Claude Sonnet 5.5
2026-09-28
Gemini 4 Carbon
2026-10-28
Gemini 4 Carbon has the higher public point estimate, 88.42 versus 81.86. Their conditional score ranges overlap. These ranges do not establish rank confidence. The page therefore keeps the decision tied to the specific documented workload.
Claude Sonnet 5.5 has the higher public coding point estimate, 83.8 to 82.2, with Supported evidence for both models. The conditional ranges do not establish rank confidence.
Gemini 4 Carbon is not ranked on the public lane for agentic tasks, so no winner is named for agentic tasks.
For the stated presets, chat costs $0.007 on Claude Sonnet 5.5 and $0.007 on Gemini 4 Carbon; repository review costs $0.13 and $0.13; the cache-heavy agent loop costs $0.16 and $0.16. Costs use the listed standard API rates.
A complete documented context-window comparison is not available.
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Terminal-Bench 4.0
Not directly comparable
HLE w/ tools
Not directly comparable
DRACO
Not directly comparable
Terminal-Bench-Science 0.1
Not directly comparable
LAB all-pass (Harvey held-out)
Not directly comparable
LAB criterion-pass (Harvey held-out)
Not directly comparable
Toolathlon Verified Pass@3
Not directly comparable
Toolathlon Verified Pass³
Not directly comparable
Toolathlon Verified avg. turns
Not directly comparable
AutomationBench (Zapier 1.0.6)
Not directly comparable
Toolathlon-Verified
Not directly comparable
Hermes Index
Not directly comparable
Hermes Bench
Not directly comparable
MazeBench
Not directly comparable
OSWorld 2.1
Not directly comparable
OSWorld 2.1 Offline
Claude Sonnet 5.5 leads this result
AutomationBench
Not directly comparable
CWE-bench v1
Not directly comparable
ExploitGym
Not directly comparable
FrontierCode 1.1 Main
Not directly comparable
CursorBench 4.0
Not directly comparable
SWE-bench Pro
Not directly comparable
SWE Multilingual
Not directly comparable
SWE Multimodal
Not directly comparable
DeepSWE
Not directly comparable
FrontierCode 1.1 Extended
Not directly comparable
ProgramBench
Not directly comparable
FrontierSWE v2
Shared sourceClaude Sonnet 5.5 leads this result
Bug Hunt Bench
Not directly comparable
FrontierCode 2.0
Not directly comparable
GraphWalks BFS 256K–1M
Not directly comparable
ExtractBench Pro
Not directly comparable
ExtractBench (LlamaIndex)
Not directly comparable
ParseBench
Not directly comparable
Chartography (no tools)
Gemini 4 Carbon leads this result
Chartography (tools)
Not directly comparable
BenchCAD Vision2Code (no tools)
Not directly comparable
BenchCAD Vision2Code (tools)
Not directly comparable
Biomedical image analysis
Not directly comparable
OfficeQA
Not directly comparable
OfficeQA Pro
Not directly comparable
Blueprint-Bench 2
Not directly comparable
HealthBench (raw)
Not directly comparable
HealthBench (length-adjusted)
Not directly comparable
HealthBench Professional (raw)
Not directly comparable
HealthBench Professional
Not directly comparable
BioMysteryBench (human-solvable)
Not directly comparable
BioMysteryBench (human-difficult)
Not directly comparable
SpatialBench Verified
Not directly comparable
SingleCellBench
Not directly comparable
Protein Design
Not directly comparable
Morphology-to-molecule matching
Not directly comparable
Medicinal chemistry
Not directly comparable
Protein Design library ranking
Not directly comparable
De novo protein-binder design
Not directly comparable
Protocols (troubleshooting)
Not directly comparable
Protocols (understanding)
Not directly comparable
HLE w/o tools
Not directly comparable
Gray Swan IPI (15 attempts)
Not directly comparable
ArXivMath Aug. 2026 (no tools)
Not directly comparable
ArXivMath Aug. 2026 (tools)
Not directly comparable
FrontierMath v2 (Tiers 1-3)
Not directly comparable
FrontierMath v2 (Tier 4)
Not directly comparable
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 6,100+ subscribers.
Last updated October 28, 2026
