Model comparison
Claude Sonnet 5 vs Gemini 3.8 Flash
Anthropic against Google, compared on context, price, and verified benchmark results.
Catalog record checked September 3, 2026Individual provider fields may changeEvidence confidence: High — see receipts below
Anthropic
Claude Sonnet 5
Balanced
Gemini 3.8 Flash
Balanced
| Specification | Claude Sonnet 5 | Gemini 3.8 Flash |
|---|---|---|
| Provider | Anthropic | |
| Tier | Balanced | Balanced |
| Context window | 1M | Winner: 1.05M |
| Max output | Winner: 128K | 66K |
| Input / 1M tokens | $2 | Winner: $0.75 |
| Output / 1M tokens | $10 | Winner: $3.75 |
| Weights | Closed | Closed |
| Parameters | Not disclosedUnverified | Not disclosedUnverified |
| Reasoning levels | adaptive | low, medium, high |
| Modalities | text, image | text, image, video, audio, pdf |
| API model id | claude-sonnet-5 | gemini-3.8-flash |
| Released | June 30, 2026 | September 2, 2026 |
| Artificial Analysis Intelligence Index (2026-08-14) | 55 | Not verifiedUnverified |
| Artificial Analysis Intelligence Index [high] (2026-09-02) | Not verifiedUnverified | 58.7 |
| Terminal-Bench 2.1 (2026-06-30) | 80.4 | Winner: 89.4 |
| Humanity's Last Exam (2026-06-30) | Winner: 57.4 | 54.9 |
| SWE-bench Verified (2026-09-01) | 79.6 | Not verifiedUnverified |
| LiveBench (2026-09-05) | 76 | Not verifiedUnverified |
Prices are USD per million tokens at standard rates, excluding batch and caching discounts. Bold indicates the better figure where one is objectively better. Values we could not confirm from the provider are shown as “Not verified” rather than estimated.
Benchmark receipts
- 2026-09-05: LiveBench official leaderboard (benchmark-owned), xhigh effortContamination-resistant general capability across reasoning, coding, math, data analysis, and language, with monthly question refreshes. Comparability: comparable with caveat — Rolling question set: observations months apart measure different question mixes. Record the measurement date and compare within ~1 month windows.
- 2026-09-02: Google official blog (vendor, HLE-Verified full set)Frontier-knowledge ceiling: extremely hard multi-domain questions written to be near-impossible without deep expertise. Comparability: comparable with caveat — Subset/tool configurations (text-only vs with-tools) differ between vendors and materially change scores; must match configuration to compare.
- 2026-09-02: Google launch eval table (transcribed by Vellum)Agentic terminal work: multi-step tasks executed in a sandboxed shell environment. Comparability: comparable with caveat — Scores across Terminal-Bench major versions (v1 vs v2) are NOT comparable; harness configuration (container, time limits) also shifts results.
- 2026-09-02: Artificial AnalysisComposite index blending reasoning, knowledge, and coding evals into one 0–100 score. Comparability: directly comparable — AA occasionally rebaselines the index scale between snapshots — a score captured on one date is only comparable to same-snapshot scores (check measuredAt).
- 2026-09-01: vals.ai SWE-bench Verified leaderboard (independent, mini-swe-agent bash-only); not published by Anthropic; a 72.7 CosmicJS claim is unverifiedReal GitHub issue resolution: does the model's patch pass the hidden tests. Comparability: comparable with caveat — Post-audit vendor claims and pre-audit scores sit on different task trust levels; scaffolding (agent harness, compute budget) also dominates results. Never aggregate across scaffolds.
- 2026-06-30: Anthropic announcement (vendor, with tools)Frontier-knowledge ceiling: extremely hard multi-domain questions written to be near-impossible without deep expertise. Comparability: comparable with caveat — Subset/tool configurations (text-only vs with-tools) differ between vendors and materially change scores; must match configuration to compare.
- 2026-06-30: Anthropic announcement table (vendor, default effort; via Vellum transcription)Agentic terminal work: multi-step tasks executed in a sandboxed shell environment. Comparability: comparable with caveat — Scores across Terminal-Bench major versions (v1 vs v2) are NOT comparable; harness configuration (container, time limits) also shifts results.
Pricing tiers: Claude Sonnet 5: Standard $2/$10 per MTok; cache reads 10% of base input. Knowledge cutoff Jan 2026. · Gemini 3.8 Flash: Intro $0.75/$3.75 per MTok through 2026-12-31, rising to $1.50/$7.50 from 2027-01-01; batch/Flex half price. Free tier available. Companion Gemini 3.8 Flash Cyber (defensive security) is restricted to the Fairwind Program with no public pricing.
Claude Sonnet 5
Claude Sonnet 5 is Anthropic's best speed/intelligence balance for high-volume agentic work at $2/$10.
Best for
- Everyday generation
- High-volume agents
- Cost-sensitive work
Watch out
A step below Opus 5 on the hardest tasks; pick Opus when quality is paramount.
Gemini 3.8 Flash
Gemini 3.8 Flash is Google's most-intelligent workhorse Flash — tied for the top of DeepSWE 1.1 with Claude Opus 5 at roughly a fifth of the cost per task, three Flash releases in six weeks.
Best for
- Agentic coding at scale
- Mid-difficulty engineering
- Multimodal pipelines
Watch out
Intro pricing doubles on 2027-01-01, and Google says it can consume more tokens than 3.7 Flash — 3.7 Flash stays available for efficiency-first workloads.
Benchmark
DeepSWE 1.1 in context
Only Gemini 3.8 Flash has a published DeepSWE 1.1 result. It is shown against the wider field, with cost per completed task alongside the score.
Local leader
Gemini 3.8 Flash [high]
74%
Rows shown
27
Highest published reasoning effort per model (not best Pass@1)
Snapshot date
2026-09-03
Mirrored from deepswe.datacurve.ai
Better is toward the top-right (higher pass rate, lower cost). X-axis is reversed to match DeepSWE’s public chart. v1.1 uses average cost / tokens / steps; v1 uses published medians.
| # | Model | Pass@1 | Cost / task | Tokens / task | Steps / task |
|---|---|---|---|---|---|
| 1 | Gemini 3.8 Flash [high] | 74% | $2.36 | 143k | 166 |
| 2 | Claude Opus 5 [max] | 74% | $11.84 | 118k | 99 |
| 3 | GPT-5.6 Sol [max] | 73% | $8.39 | 60k | 61 |
| 4 | Claude Fable 5 [max] | 70% | $21.63 | 119k | 88 |
| 5 | GPT-5.6 Terra [max] | 70% | $4.95 | 72k | 76 |
| 6 | GLM 5.3 [max] | 69% | $3.99 | 80k | 124 |
| 7 | Kimi K3 [max] | 69% | $4.65 | 82k | 98 |
| 8 | GPT-5.6 Luna [max] | 67% | $3.03 | 73k | 102 |
| 9 | GPT-5.5 [xhigh] | 67% | $7.23 | 46k | 82 |
| 10 | Grok 4.6 [xhigh] | 67% | $5.50 | 71k | 87 |
| 11 | Gemini 3.7 Flash [high] | 65% | $2.18 | 107k | 125 |
| 12 | GLM 5.3 Flash [max] | 63% | $0.48 | 73k | 123 |
| 13 | DeepSeek V4-Pro [max] | 63% | $0.24 | 106k | 155 |
| 14 | Claude Opus 4.8 [max] | 59% | $13.22 | 135k | 120 |
| 15 | Qwen3.8-Max [xhigh] | 58% | $3.73 | 95k | 111 |
| 16 | Muse Spark 1.2 [xhigh] | 55% | $3.70 | 99k | 101 |
| 17 | Claude Sonnet 5 [max] | 54% | $26.40 | 214k | 268 |
| 18 | Grok 4.5 [high] | 54% | $2.42 | 36k | 61 |
| 19 | DeepSeek V4-Flash [max] | 53% | $0.10 | 108k | 153 |
| 20 | Muse Spark 1.1 [xhigh] | 53% | $2.36 | 74k | 96 |
| 21 | GPT-5.4 [xhigh] | 52% | $5.65 | 71k | 70 |
| 22 | Gemini 3.6 Flash [high] | 47% | $4.42 | 96k | 117 |
| 23 | GLM 5.2 [max] | 44% | $3.92 | 78k | 129 |
| 24 | Gemini 3.5 Flash [high] | 36% | $3.45 | 76k | 105 |
| 25 | Kimi K2.7 Code | 31% | $2.82 | 59k | 149 |
| 26 | Claude Sonnet 4.6 [high] | 30% | $5.52 | 76k | 134 |
| 27 | Gemini 3.1 Pro [high] | 12% | $2.14 | 28k | 76 |
DeepSWE “Best” picks the highest published reasoning effort per model (not the highest pass rate). Small gaps may not be statistically meaningful — confirm on deepswe.datacurve.ai.
When the cheaper one wins
Gemini 3.8 Flash is cheaper on output at $3.75 per million tokens against $10 for Claude Sonnet 5 — about 2.7×. Use the cheaper tier for classification, extraction, summarisation, and any task where the expensive model’s extra score does not change the accepted output. The expensive one only pays if your hardest task actually fails on the cheap tier. These are standard-tier API rates, excluding batch and cache discounts.
Run the model pickerEvidence confidence: High
How strong and complete the evidence behind this comparison is — not a prediction of which model is better.
- Pricing verified on both sides — Input and output rates are verified for both models.
- 4/5 core specs verified on both sides — Not published for at least one side: parameter count.
- 3 shared named benchmarks with differing scores — Measured on: Artificial Analysis Intelligence Index, Terminal-Bench 2.1, Humanity's Last Exam.
- Verified within the last 90 days — Newest catalog check was 8 days ago.
- Both models carry source citations — Each side has at least two catalog sources on record.
Source receipts
Each catalog figure was checked against the provider or an independent second source on the date shown.
- Claude Sonnet 5: Anthropic — Claude Sonnet 5 (accessed 2026-08-29)
- Claude Sonnet 5: Anthropic news — Claude Sonnet 5 (accessed 2026-08-29)
- Gemini 3.8 Flash: Google — Gemini 3.8 Flash and 3.8 Flash Cyber (accessed 2026-09-03)
- Gemini 3.8 Flash: Gemini API pricing (gemini-3.8-flash $0.75/$3.75 intro) (accessed 2026-09-03)
- Gemini 3.8 Flash: DeepSWE 1.1 leaderboard — gemini-3.8-flash [high] 73.8% at $2.36/task (accessed 2026-09-03)
Related comparisons
- Claude Opus 5 vs Gemini 3.8 Flash
- Gemini 3.8 Flash vs GPT-5.6 Sol
- Amazon Nova 2 Pro vs Claude Sonnet 5
- Amazon Nova 2 Pro vs Gemini 3.8 Flash
- Claude Fable 5.1 vs Claude Sonnet 5
- Claude Fable 5 vs Claude Sonnet 5
Diving deeper on one model? Claude Sonnet 5 · Gemini 3.8 Flash
Common questions
Claude Sonnet 5 vs Gemini 3.8 Flash
Answered from the verified figures on this page rather than general guidance.
Is Claude Sonnet 5 or Gemini 3.8 Flash cheaper for input?
Gemini 3.8 Flash is cheaper at $0.75 per million input tokens, against $2 for Claude Sonnet 5 — roughly 2.7× the price. Output tokens usually dominate a real bill, so weigh the output rate more heavily than the input rate. These are base rates; Claude Sonnet 5 has tiered pricing: Standard $2/$10 per MTok; cache reads 10% of base input. Knowledge cutoff Jan 2026. Gemini 3.8 Flash has tiered pricing: Intro $0.75/$3.75 per MTok through 2026-12-31, rising to $1.50/$7.50 from 2027-01-01; batch/Flex half price. Free tier available. Companion Gemini 3.8 Flash Cyber (defensive security) is restricted to the Fairwind Program with no public pricing.
Is Claude Sonnet 5 or Gemini 3.8 Flash cheaper for output?
Gemini 3.8 Flash is cheaper at $3.75 per million output tokens, against $10 for Claude Sonnet 5 — roughly 2.7× the price. Output tokens usually dominate a real bill, so weigh the output rate more heavily than the input rate. These are base rates; Claude Sonnet 5 has tiered pricing: Standard $2/$10 per MTok; cache reads 10% of base input. Knowledge cutoff Jan 2026. Gemini 3.8 Flash has tiered pricing: Intro $0.75/$3.75 per MTok through 2026-12-31, rising to $1.50/$7.50 from 2027-01-01; batch/Flex half price. Free tier available. Companion Gemini 3.8 Flash Cyber (defensive security) is restricted to the Fairwind Program with no public pricing.
Which has the larger context window, Claude Sonnet 5 or Gemini 3.8 Flash?
Gemini 3.8 Flash accepts 1.05M tokens against 1M for Claude Sonnet 5. This only matters if you routinely send very long documents or large codebases.
Do Claude Sonnet 5 and Gemini 3.8 Flash support the same reasoning levels?
Claude Sonnet 5 exposes adaptive, while Gemini 3.8 Flash exposes low, medium, high.
Should I use Claude Sonnet 5 or Gemini 3.8 Flash?
Both sit in the balanced tier, so the choice usually comes down to price and context rather than capability. Claude Sonnet 5 suits everyday generation; Gemini 3.8 Flash suits agentic coding at scale.
Next step
Choosing between them
The verified figures that separate this pair, computed from the catalog rather than restated boilerplate.
- Input price: Gemini 3.8 Flash costs $0.75 per 1M tokens versus $2 for Claude Sonnet 5 — a 2.7x difference at the headline tier.
- Context: Gemini 3.8 Flash takes 1.05M against 1M for Claude Sonnet 5 — only decisive if your prompts approach the smaller window.
- Measured capability: Gemini 3.8 Flash leads Artificial Analysis Intelligence Index 58.7 to 55 (measured 2026-08-14).
Choosing a harness rather than a model — Cursor, Copilot, Claude Code, Windsurf, or Lovable? Compare agentic harnesses · Latest releases.