Across 24 comparable benchmarks, Gemini 3.8 Flash has better reported results on 23 and worse on 1. 3 further shared results are listed below, each with the reason it is not counted.
Moderate evidence24 of 27 shared results countedHow we compare
Price basis: Current effective rates. The price section below shows the applicable periods.
| Fact | Haiku 4.5 | Gemini 3.8 Flash |
|---|---|---|
| Context window | 200K tokens | 1M tokens |
| Input / 1M | $1.00 | $0.750 |
| Output / 1M | $5.00 | $3.75 |
| Blended (4:1) | $1.15 | $0.432 |
| Status | Available | Available |
| Released | October 15, 2025 | September 2, 2026 |
Blended is 4:1 input to output with 90% cache hits at listed rates β set your own mix.
Mean of 15 shared benchmarks grouped by capability on a 0β100 scale. How averages are built
Gemini 3.8 Flash ahead on 3 of 3 counted
Gemini 3.8 Flash ahead on 1 of 1 counted
Gemini 3.8 Flash ahead on 3 of 4 counted
Gemini 3.8 Flash ahead on 2 of 2 counted
Gemini 3.8 Flash ahead on 9 of 9 counted
No shared comparable results for Mathematics, Computer & tool use, Research & search, Long context, Cybersecurity.
Published scores, with matching modes where available. Gaps show Haiku 4.5 relative to Gemini 3.8 Flash. Reading the results
| Benchmark | Haiku 4.5 | Gemini 3.8 Flash | Gap |
|---|---|---|---|
| Code MigrationMaintenance & Optimization Β· Hidden-test pass rate (%) | 7.5Thinking Β· tools | 36.5β Thinking High Β· tools | β29.0 pts#67 vs #37 of 79 |
| SWE-Bench Pro V2 HardRepository Engineering Β· Resolve Rate (%, pass@1) | 25.5Extra-High Β· tools | 58.8β Thinking High Β· tools | β33.3 pts#21 vs #18 of 21 |
| Vibe Code Bench v1.1Code Generation & Editing Β· Pass rate (%) | 11.4Thinking Β· tools | 78.7β Thinking High Β· tools | β67.3 pts#102 vs #35 of 121 |
| ARC-AGI-1Abstract Generalization Β· Score (%) | 47.7Extended Β· no tools | 98.5β Thinking High Β· no tools | β50.8 pts#67 vs #1 of 87 |
| ARC-AGI-2Abstract Generalization Β· Score (% solved); cost per task (USD)Not counted Β· Speed and cost telemetry is reported, not scored | 4.5Extended Β· no tools | 89.2β Thinking High Β· no tools | 0.05Γ#70 vs #12 of 80 |
| Vending-Bench 2Logical Planning Β· Money Balance (USD)Not counted Β· Speed and cost telemetry is reported, not scored | 459Unspecified Reasoning Β· tools | 5094Unspecified Reasoning Β· tools | 0.09Γ#58 vs #38 of 64 |
| GPQA DiamondScientific Reasoning Β· Accuracy | 73.3Extended Β· no tools | 95.3β Thinking High Β· no tools | β22.0 pts#126 vs #7 of 200 |
| SciCodeScientific Computing Β· Score | 42.2Thinking Β· no tools | 56.6β Thinking High Β· no tools | 0.75Γ#72 vs #24 of 98 |
| MedCodeClinical Workflows Β· Accuracy (%) | 32.7Thinking Β· tools | 48.1β Thinking High Β· tools | β15.5 pts#81 vs #30 of 97 |
| MedScribeClinical Workflows Β· Rubric pass rate (%) | 85.2Thinking Β· tools | 84.5β Thinking High Β· tools | +0.7 pts#30 vs #34 of 99 |
| GDP.pdfDocuments & Charts Β· Task pass rate (%) | 3.8Thinking Β· no tools | 22.8β Thinking Medium Β· no tools | β19.0 pts#78 vs #22 of 96 |
| MMMU-ProVisual Understanding Β· Accuracy | 58.6Thinking Β· no tools | 85.6β Thinking High Β· no tools | β27.0 pts#109 vs #7 of 136 |
| AA-AnalystAgentData Analysis Β· pass^5 (%) | 15.0Thinking Β· tools + web | 46.3β Thinking High Β· tools + web | β31.3 pts#32 vs #17 of 39 |
| AutomationBenchOffice & Business Β· Pass Rate | 0.46Reported bestΒ· zapier-private-heldout | 29.7β Thinking Medium Β· toolsΒ· zapier-private-heldout | β29.2 pts#62 vs #35 of 64 |
| EMB (Excel Modeling)Finance Β· Accuracy (%) | 22.7Thinking Β· tools | 72.2β Thinking High Β· tools | β49.5 pts#72 vs #10 of 78 |
| Finance Agent v2Finance Β· Score | 31.0Thinking Β· tools | 61.4Thinking Β· tools | 0.51Γ#68 vs #2 of 75 |
| Harvey's Legal Agent BenchmarkLegal Β· Accuracy (%) | 0.83Thinking Β· tools | 10.0β Thinking High Β· tools | β9.2 pts#41 vs #13 of 52 |
| Legal Research BenchLegal Β· Score | 10.6Thinking Β· tools | 38.9β Thinking High Β· tools | 0.27Γ#81 vs #39 of 91 |
| Public Benefits Bench v1.1Service Workflows Β· Criterion pass rate (%) | 54.3Unspecified Reasoning Β· tools | 65.3β Thinking High Β· tools | β11.0 pts#42 vs #22 of 50 |
| SAGEService Workflows Β· Accuracy (%) | 31.8Thinking Β· tools | 35.1β Thinking High Β· tools | β3.2 pts#71 vs #66 of 82 |
| ΟΒ³-BankingService Workflows Β· Score | 9.3Thinking Β· tools | 45.8β Thinking Medium Β· tools | 0.20Γ#95 vs #13 of 120 |
| HLEKnowledge Exams Β· Accuracy | 9.7Extended Β· no tools | 44.5β Reported best | β34.8 pts#129 vs #66 of 151 |
| AA Intelligence Index v5.0.1Capability Indices Β· Index (0-100) | 7.4Thinking Β· tools | 30.8β Thinking High Β· tools | 0.24Γ#54 vs #29 of 62 |
| ECICapability Indices Β· ECI score (capability index, higher is better) | 142Reported best | 157Reported best | 0.91Γ#90 vs #15 of 168 |
| LiveBenchCross-capability Suites Β· Accuracy | 61.3Thinking Β· no tools Β· 64K | 75.8β Thinking High Β· no tools | β14.5 pts#82 vs #30 of 106 |
| LiveBench 2.0Cross-capability Suites Β· Overall score (equal-weight mean of 7 suite scores, 0-100) | 46.0Max Β· task tools | 67.3β Thinking High Β· tools | 0.68Γ#48 vs #29 of 51 |
| Vals IndexCapability IndicesNot counted Β· This metric has no better direction | 22.9Thinking Β· tools | 62.3β Thinking High Β· tools | β |
Maintenance & Optimization Β· Hidden-test pass rate (%)
Repository Engineering Β· Resolve Rate (%, pass@1)
Code Generation & Editing Β· Pass rate (%)
Abstract Generalization Β· Score (%)
Abstract Generalization Β· Score (% solved); cost per task (USD)
Not counted Β· Speed and cost telemetry is reported, not scored
Logical Planning Β· Money Balance (USD)
Not counted Β· Speed and cost telemetry is reported, not scored
Scientific Reasoning Β· Accuracy
Scientific Computing Β· Score
Clinical Workflows Β· Accuracy (%)
Clinical Workflows Β· Rubric pass rate (%)
Data Analysis Β· pass^5 (%)
Office & Business Β· Pass Rate
Finance Β· Accuracy (%)
Finance Β· Score
Legal Β· Accuracy (%)
Legal Β· Score
Service Workflows Β· Criterion pass rate (%)
Service Workflows Β· Accuracy (%)
Service Workflows Β· Score
Knowledge Exams Β· Accuracy
Capability Indices Β· Index (0-100)
Capability Indices Β· ECI score (capability index, higher is better)
Cross-capability Suites Β· Accuracy
Cross-capability Suites Β· Overall score (equal-weight mean of 7 suite scores, 0-100)
Capability Indices
Not counted Β· This metric has no better direction
A dash means no result is available.
Both models publish peak and off-peak prices. Each period is compared separately using the same token mix; hours follow each providerβs schedule.
Drag the handle above the bar for cache hits, the one below it for the input : output mix Β· 5% steps
Blended price = effective input rate at 90% cache hits Γ 80% + output rate Γ 20%
Haiku 4.5 costs33%morethan Gemini 3.8 Flash per million tokens.Using current effective rates at this token mix.
Haiku 4.5: Standard pricing. Anthropic official pricing (platform.claude.com/docs/en/about-claude/pricing) verified 2026-11-13
Gemini 3.8 Flash: Listed text rates; pricing tier not confirmed.
| Price / 1M tokens | Haiku 4.5 | Gemini 3.8 Flash | Ratio |
|---|---|---|---|
| Input | $1.00 | $0.75 | 1.33Γ |
| Output | $5.00 | $3.75 | 1.33Γ |
| Cached input | $0.10 | $0.075 | 1.33Γ |
| Cache write | β | β | β |
| Blended | $1.15 | $0.864 | 1.33Γ |
Haiku 4.5 costs33%lessthan Gemini 3.8 Flash per million tokens.Using standard list rates after promotions end at this token mix.
Haiku 4.5: Standard pricing. Anthropic official pricing (platform.claude.com/docs/en/about-claude/pricing) verified 2026-11-13
Gemini 3.8 Flash: Standard pricing. Google Gemini API official pricing (ai.google.dev/gemini-api/docs/pricing) verified 2026-10-21; paid tier; standard tier; scheduled price starting 2027-01-01
| Price / 1M tokens | Haiku 4.5 | Gemini 3.8 Flash | Ratio |
|---|---|---|---|
| Input | $1.00 | $1.50 | 0.67Γ |
| Output | $5.00 | $7.50 | 0.67Γ |
| Cached input | $0.10 | $0.15 | 0.67Γ |
| Cache write | β | β | β |
| Blended | $1.15 | $1.73 | 0.67Γ |
Estimates use the selected token mix. Cache writes, tool fees and long-context surcharges are excluded.
Reported cost per task, paired with the score from the same run. Settings and sources may differ; costs do not affect the verdict.
Haiku 4.5 relative to Gemini 3.8 Flash
No matching recorded settings for a cost summary.
Summary: 0 of 4 shared benchmarks have matching recorded settings and sources. The average is a geometric mean of 0 positive-cost ratios, one equal weight per benchmark. This benchmark set can include related tests and indexes; it is not a workload cost estimate.
Settings or sources differ
Settings or sources differ
Settings or sources differ
Settings or sources differ
Settings or sources differ
Settings or sources differ
Settings or sources differ
Settings or sources differ
Settings or sources differ or are not recorded; interpret the comparison with care.
Haiku 4.5
Gemini 3.8 Flash
Settings or sources differ or are not recorded; interpret the comparison with care.
Haiku 4.5
Gemini 3.8 Flash
Settings or sources differ or are not recorded; interpret the comparison with care.
Haiku 4.5
Gemini 3.8 Flash
Settings or sources differ or are not recorded; interpret the comparison with care.
Haiku 4.5
Gemini 3.8 Flash
| Benchmark | Haiku 4.5 | Gemini 3.8 Flash |
|---|---|---|
| AA-LCR | Not reported | |
| Agent Arena | Not reported | $0.7652 / task Median Β· Net improvement index (percentage points vs. average model): 2.3 Thinking High (With Tools + Internet) |
| Agent Arena Index | Not reported | $0.7486 / task Median Β· Agent Arena Index rating (index points): 1118 Thinking High (With Tools + Internet) |
| ARC-AGI-2 | Not reported | $0.3997 / task Average Β· ARC-AGI-2 Score: 89.2 Thinking High (No Tools) $0.1545 / task Average Β· ARC-AGI-2 Score: 77.5 Thinking Low (No Tools) $0.2478 / task Average Β· ARC-AGI-2 Score: 82.9 Thinking Medium (No Tools) |
| CursorBench 4.0 | Not reported | $4.70 / task Average Β· Task score: 39.6 Thinking High (With Tools) Β· cursorbench-4-0-cursor-agent $4.06 / task Average Β· Task score: 37.3 Thinking Medium (With Tools) Β· cursorbench-4-0-cursor-agent |
| DeepSWE | Not reported | |
| FrontierSWE v2 | Not reported | |
| Public Benefits Bench v1.1 | $0.2245 / task Average Β· Criterion pass rate (%): 54.3 Unspecified Reasoning (With Tools) | Not reported |
| Terminal-Bench 4.0 | Not reported | $8.65 / task Average Β· Resolution rate (%): 19.7 Unspecified Reasoning (With Tools) Β· vals-mini-swe |
| Terminal-Bench 4.1 | Not reported | |
| Vals Index v2.1 | Not reported | |
| Vals Index v2.2 | Not reported | |
| Vibe Code Bench 1-100 | Not reported | |
| WeirdML v3 | Not reported | $3.9441 / task Average Β· Official score (0-1; higher is better): 0.074 Thinking High (With Tools) |
| Attribute | Haiku 4.5 | Gemini 3.8 Flash |
|---|---|---|
| Availability | Available | Available |
| Provider | Anthropic | Google Deep Mind |
| Model type | Multimodal model | Multimodal model |
| First release | October 15, 2025 | September 2, 2026 |
| Context window | 200K tokens | 1M tokens |
| Max output | 64K tokens | 64K tokens |
| Architecture | Undisclosed | Undisclosed |
| Documented capabilities | Reasoning, Multilingual | Reasoning, Coding, Function calling, Tool use, Structured output, Computer use, Web search, Long context |
| Input modalities | Text, Image | Text, Image, Audio, Video |
| Output modalities | Text | Text |
| Weights | Closed / API only | Closed / API only |
| Commercial use | Proprietary | Proprietary |
These are reported-score comparisons, not controlled head-to-head tests.
What people found comparing these models, or using either one. Comments also appear on the models' own pages.
Used Haiku 4.5 or Gemini 3.8 Flash? Sign in with Google to share what worked, what did not, and which API or setup you tested.
No community comments on either model yet.
Alternatives with the most benchmark overlap for this matchup.