Agentic work
Tool use, computer use, and multi-step task completion
Claude Sonnet 5.5
Claude Sonnet 5.5 leads on the public agentic lane, 66.1 to 57.9, with Supported evidence for both models, although the 90% intervals overlap.
Updated September 28, 2026. Rank says Claude Sonnet 5.5 is ahead. Price, access, and your workload can each overturn that. Public scores include evidence status and uncertainty.
Both of these models will change. Get the price, version and retirement notices for the pair, each with its source. Follow model changes
Claude Sonnet 5.5 has the higher public score estimate, 80.49 versus 67.53, but the 90% score intervals overlap. Treat that as a lead, not a settled winner. 5 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.
Tool use, computer use, and multi-step task completion
Claude Sonnet 5.5
Claude Sonnet 5.5 leads on the public agentic lane, 66.1 to 57.9, with Supported evidence for both models, although the 90% intervals overlap.
1K fresh input + 500 output tokens
Gemini 3.7 Flash
Gemini 3.7 Flash has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
200K cached + 20K fresh input + 10K output tokens
Gemini 3.7 Flash
Gemini 3.7 Flash has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
50K fresh input + 3K output tokens
Gemini 3.7 Flash
Gemini 3.7 Flash has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Code generation, repair, and software-engineering tasks
Not enough matched evidence
Claude Sonnet 5.5 is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like.
Prompts that approach the documented context limit
No clear pick
The documented context windows are equal.
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.
Directional only · BenchAlign v5.7
Claude Sonnet 5.5 scores higher, but at least one score rests on Estimated evidence or a different benchmark set. Directional only, no winner.
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.
1 category rests on Estimated evidence or different evidence sets. Those rows remain visible for coverage context but do not name a winner.
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 v5.7 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 3.7 Flash | Basis | Reading |
|---|---|---|---|---|
| Agentic | 66.1Supported · #10/111 | 57.9Supported · #21/111 | Like-for-likeBenchAlign v5.7 lane · 11 vs 7 public rows | Claude Sonnet 5.5 leads · intervals overlap |
| Knowledge | 80.9Supported · #6/160 | 71.0Supported · #11/160 | Like-for-likeBenchAlign v5.7 lane · 16 vs 6 public rows | Claude Sonnet 5.5 leads · intervals overlap |
| Coding | 79.6Estimated · #3/136 | 59.4Supported · #18/136 | Directional onlyBenchAlign v5.7 lane · 9 vs 6 public rows | Directional only |
| Reasoning | 79.2#7/27 | 77.9Unranked · 5 rankable rows | Not comparableProvisional lane · 0 vs 1 weighted rows | Not comparable |
| Multimodal | 83.3Unranked · 7 rankable rows | 83.6#10/50 | Not comparableProvisional lane · 1 vs 1 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 | Not ranked | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
Ranks count the models scored in each category’s lane, so the agentic, coding, and knowledge denominators (BenchAlign v5.7) 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
Gemini 3.7 Flash has the lower modeled cost
Costs use the listed standard API rates.
50K fresh input + 3K output tokens
Gemini 3.7 Flash has the lower modeled cost
Costs use the listed standard API rates.
200K cached + 20K fresh input + 10K output tokens
Gemini 3.7 Flash has the lower modeled cost
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 3.7 Flash
Claude Sonnet 5.5
claude-sonnet-5-5
Anthropic Claude Sonnet 5.5 model documentationGemini 3.7 Flash
gemini-3.7-flash
Google Gemini 3.7 Flash API documentationA missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
Claude Sonnet 5.5
$0.2 per 1M cached input tokens
Claude API pricingGemini 3.7 Flash
$0.075 per 1M cached input tokens
Google Gemini API pricingClaude Sonnet 5.5
Gemini 3.7 Flash
text, image, video, audio, pdf
Google Gemini 3.7 Flash API documentationClaude Sonnet 5.5
Gemini 3.7 Flash
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 3.7 Flash
Generally Available · Gemini API, Google AI Studio, Gemini App - Spark, Gemini Enterprise App, Gemini Enterprise Agent Platform, Google Antigravity
Google DeepMind Gemini 3.7 Flash model cardClaude Sonnet 5.5
Reasoning
Gemini 3.7 Flash
Reasoning
Claude Sonnet 5.5
Proprietary
Gemini 3.7 Flash
Proprietary
Claude Sonnet 5.5
Proprietary
Gemini 3.7 Flash
Proprietary
Claude Sonnet 5.5
2026-09-28
Gemini 3.7 Flash
2026-08-13
Claude Sonnet 5.5 has the higher public score estimate, 80.49 versus 67.53, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.
Claude Sonnet 5.5 scores higher for coding on the public lane, 79.6 to 59.4. Claude Sonnet 5.5 is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like. BenchLM does not name a winner for coding from a directional reading; read the shared benchmark rows directly and test the models on the same task set.
Claude Sonnet 5.5 leads the public agentic tasks lane, 66.1 to 57.9, with Supported evidence for both models, although the 90% intervals overlap.
For the stated presets, chat costs $0.007 on Claude Sonnet 5.5 and $0.00263 on Gemini 3.7 Flash; repository review costs $0.13 and $0.04875; the cache-heavy agent loop costs $0.18 and $0.0675. Costs use the listed standard API rates.
Both models list the same context window, 1M.
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
Terminal-Bench 2.1
Not directly comparable
Terminal-Bench 3.0
Not directly comparable
AutomationBench
Not directly comparable
OSWorld 2.0
Not directly comparable
Agents' Last Exam
Not directly comparable
Terminal-Bench 2.1 (Vals)
Not directly comparable
ApprenticeBench
Not directly comparable
FrontierCode 1.1 Main
Claude Sonnet 5.5 leads this result
cursorBench40
Not directly comparable
SWE-bench Pro
Not directly comparable
SWE Multilingual
Not directly comparable
SWE Multimodal
Not directly comparable
DeepSWE
Claude Sonnet 5.5 leads this result
FrontierCode 1.1 Extended
Not directly comparable
ProgramBench
Not directly comparable
FrontierSWE v2
Claude Sonnet 5.5 leads this result
Terminal-Bench 2.1
Not directly comparable
LiveCodeBench (Vals)
Not directly comparable
SWE-bench (Vals)
Not directly comparable
Chartography (no tools)
Not directly comparable
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
CharXiv w/o tools
Not directly comparable
CharXiv
Not directly comparable
LVBench
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)
Claude Sonnet 5.5 leads this result
BioMysteryBench (human-difficult)
Claude Sonnet 5.5 leads this result
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
HLE-Verified
Not directly comparable
LABBench2
Not directly comparable
GPQA Diamond (Vals)
Not directly comparable
MMLU-Pro (Vals)
Not directly comparable
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.
Last updated September 28, 2026