Claude Sonnet 4.6 vs Gemini 3.5 Flash
Claude Sonnet 4.6 and Gemini 3.5 Flash are Anthropic and Google frontier-grade production workhorses with about 1M-token context windows. Gemini launched in May 2026 with lower standard pricing at $1.50/$9 per 1M tokens, higher measured throughput, native audio input, and strong agentic and multimodal rows. Claude Sonnet 4.6 costs $3/$15 on the standard tier, but returns the first token much faster in the researched speed data and keeps a small coding edge on SWE-bench Verified and HumanEval. The decision is cost and throughput versus first-token latency, coding confidence, computer use, and enterprise provider reach.
Pick Gemini 3.5 Flash for cost-sensitive production APIs, batch volume, agentic tool loops, and multimodal pipelines: it is 50% cheaper on standard input, 40% cheaper on standard output, has the lower Batch/Flex tier, and leads MCP-Atlas at 83.6% versus 61.3%. Pick Claude Sonnet 4.6 when code correctness, low first-token latency, computer use, or enterprise provider reach matter more: it leads SWE-bench Verified at 79.6% versus 78%, HumanEval at 98% versus 92%, and the researched speed data shows 1.48s TTFT versus 18.55s for Gemini. For async jobs, Claude's $1.50/$7.50 batch tier makes it more competitive, but Gemini Batch/Flex remains cheaper at $0.75/$4.50.
Decision scorecard
Local evidence first| Signal | Claude Sonnet 4.6 | Gemini 3.5 Flash |
|---|---|---|
| Best for | reasoning-heavy apps, multimodal apps, and tool-calling agents | reasoning-heavy apps, multimodal apps, and tool-calling agents |
| Decision fit | Coding, RAG, and Agents | Coding, RAG, and Agents |
| Context window | 1m | 1.05m |
| Cheapest output | $15/1M tokens | $9/1M tokens |
| Provider routes | 6 tracked | 4 tracked |
| Shared benchmarks | SWE-bench Verified leader | 9 shared |
Decision tradeoffs
- Claude Sonnet 4.6 holds a shared-benchmark lead on SWE-bench Verified, ahead by 1.6 points.
- Claude Sonnet 4.6 has broader tracked provider coverage for fallback and procurement flexibility.
- Claude Sonnet 4.6 uniquely exposes Computer use and Parallel agents in local model data.
- Local decision data tags Claude Sonnet 4.6 for Coding, RAG, and Agents.
- Gemini 3.5 Flash holds a shared-benchmark lead on Google-Proof Q&A, ahead by 2.3 points.
- Gemini 3.5 Flash has the larger context window for long prompts, retrieval packs, or transcript analysis.
- Gemini 3.5 Flash has the lower cheapest tracked output price at $9/1M tokens.
- Local decision data tags Gemini 3.5 Flash for Coding, RAG, and Agents.
Monthly cost at traffic
Estimate token spend from the cheapest tracked input and output route or tier on this page.
Claude Sonnet 4.6
$6,150
Cheapest tracked route/tier: OpenRouter
Gemini 3.5 Flash
$3,450
Cheapest tracked route/tier: Google AI Studio
Estimated monthly gap: $2,700. Batch, cache, alternate speed tiers, and negotiated pricing are excluded from this local estimate.
Switch friction
- Provider overlap exists on GCP Vertex AI, Vercel AI Gateway, and OpenRouter; start route-level A/B tests there.
- Gemini 3.5 Flash is $6/1M tokens lower on cheapest tracked output pricing before cache, batch, or negotiated discounts.
- Check replacement coverage for Computer use and Parallel agents before moving production traffic.
- Provider overlap exists on OpenRouter, GCP Vertex AI, and Vercel AI Gateway; start route-level A/B tests there.
- Claude Sonnet 4.6 is $6/1M tokens higher on cheapest tracked output pricing, so quality gains need to justify the spend.
- Claude Sonnet 4.6 adds Computer use and Parallel agents in local capability data.
Specs
| Specification | ||
|---|---|---|
| Released | 2026-02-17 | 2026-05-19 |
| Context window | 1m | 1.05m |
| Parameters | — | — |
| Architecture | Decoder Only | Decoder Only |
| License | Proprietary | Proprietary |
| Openness | Proprietary | Proprietary |
| Weights | Not released | Not released |
| Code | Unknown | Unknown |
| Commercial use | Commercial use: conditional | Commercial use: conditional |
| Knowledge cutoff | 2025-08 | 2025-01 |
Pricing and availability
| Pricing attribute | Claude Sonnet 4.6 | Gemini 3.5 Flash |
|---|---|---|
| Input price | $3/1M tokens | $1.50/1M tokens |
| Output price | $15/1M tokens | $9/1M tokens |
| Providers |
Capabilities
| Capability | Claude Sonnet 4.6 | Gemini 3.5 Flash |
|---|---|---|
| Vision | Yes | Yes |
| Multimodal | Yes | Yes |
| Reasoning | Yes | Yes |
| Function calling | Yes | Yes |
| Tool use | Yes | Yes |
| Structured outputs | Yes | Yes |
| Code execution | Yes | Yes |
| IDE integration | No | No |
| Computer use | Yes | No |
| Parallel agents | Yes | No |
Benchmarks
| Benchmark | Claude Sonnet 4.6 | Gemini 3.5 Flash |
|---|---|---|
| SWE-bench Verified | 79.6 | 78.0 |
| Google-Proof Q&A | 89.9 | 92.2 |
| Massive Multi-discipline Multimodal Understanding | 83.6 | 83.6 |
| Humanity's Last Exam | 33.2 | 40.2 |
| HumanEval | 98.0 | 92.0 |
| MCP-Atlas | 61.3 | 83.6 |
| ARC-AGI-2 | 58.3 | 72.1 |
| MMMU Pro | 75.6 | 88.3 |
| CursorBench | 49.0 | 49.8 |
Deep dive
The cost split is the most durable first filter. Gemini 3.5 Flash lists $1.50/M input and $9/M output on the standard Google AI Studio tier, while Claude Sonnet 4.6 lists $3/M input and $15/M output on Anthropic. Gemini also has a lower Batch/Flex tier at $0.75/M input and $4.50/M output, while Claude's Batch API is $1.50/M input and $7.50/M output.
Latency and throughput point in different directions. The researched speed row shows Gemini at roughly 284 tokens per second, about 6x Claude's 44 tokens per second, which favors long generations and bulk jobs. The same source shows Claude at 1.48s time to first token versus Gemini at 18.55s, which favors real-time chat interfaces where users wait for the first visible word.
The benchmark story is mixed and needs harness labels. Claude has the cleaner coding edge on SWE-bench Verified and HumanEval, while Gemini leads GPQA, MCP-Atlas, ARC-AGI-2, and multimodal rows. The Terminal-Bench rows are not clean same-version evidence because Claude's 59.1% row is Terminal-Bench 2.0 and Gemini's 76.2% row is Terminal-Bench 2.1, so this page keeps that caveat out of the shared benchmark table.
MMMU-Pro and HLE also need variant discipline. Gemini has an 88.27% Vals.ai MMMU-Pro row and an 83.6% official Google row; Claude has a 75.6% with-tools Anthropic row and a 74.5% no-tools variant in the handoff. Claude's HLE seed row is 33.2% without tools, while a higher with-tools figure appears in secondary sources, so do not merge those into one universal score.
Capabilities divide by workflow surface. Claude Sonnet 4.6 has first-party computer use and broader Anthropic tool packaging for OS automation. Gemini 3.5 Flash has native audio input, video-aware multimodal positioning, Google Search grounding, and stronger agentic-tool benchmark evidence for MCP-style loops.
Provider reach can override token math. Claude Sonnet 4.6 is tracked across Anthropic API, AWS Bedrock, Google Vertex AI, Microsoft Foundry, OpenRouter, and Vercel AI Gateway, which helps regulated or multi-cloud buyers. Gemini 3.5 Flash is strongest when the team is comfortable with Google AI Studio or Vertex AI as the primary route, with aggregator coverage through OpenRouter and Vercel AI Gateway.
Freshness matters for knowledge tasks. The datapack corrected Claude Sonnet 4.6 to an August 2025 reliable knowledge cutoff while noting January 2026 training data; Gemini 3.5 Flash is tracked with a January 2025 cutoff. For time-sensitive work, both models should be paired with retrieval or search grounding instead of treated as authoritative standalone knowledge bases.
FAQ
Which is cheaper, Claude Sonnet 4.6 or Gemini 3.5 Flash?
Gemini 3.5 Flash is cheaper on the tracked Google AI Studio standard and Batch/Flex tiers. Standard pricing is $1.50/M input and $9/M output for Gemini versus $3/M input and $15/M output for Claude. Batch is $0.75/M input and $4.50/M output for Gemini versus $1.50/M input and $7.50/M output for Claude.
Which is better for coding, Claude Sonnet 4.6 or Gemini 3.5 Flash?
Claude Sonnet 4.6 is the safer coding pick in this pair because it leads the tracked SWE-bench Verified row at 79.6% versus 78% and HumanEval at 98% versus 92%. Gemini is still credible for coding, but its Google-official coding row in the handoff is SWE-bench Pro, which is a different benchmark from SWE-bench Verified.
Which is better for agents and tool use?
Gemini 3.5 Flash has the stronger sourced agentic-tool benchmark signal, led by MCP-Atlas at 83.6% versus Claude's 61.3%. Claude is better when the agent needs first-party computer use or Anthropic-specific tool surfaces; Gemini is better when throughput, Google grounding, and MCP-style tool loops are the main constraints.
Which model is faster?
It depends on the speed metric. Gemini 3.5 Flash has much higher researched throughput at about 284 tokens per second versus Claude at 44 tokens per second. Claude Sonnet 4.6 has much lower researched time to first token at 1.48s versus Gemini at 18.55s, which is better for interactive streaming UX.
Do both models have 1M context windows?
Yes. Claude Sonnet 4.6 is tracked at 1,000,000 tokens and Gemini 3.5 Flash at 1,048,576 tokens. Treat the windows as effectively equal for product planning, then verify tokenization and provider-specific limits on the exact route you plan to use.
Where can I run Claude Sonnet 4.6 and Gemini 3.5 Flash?
Claude Sonnet 4.6 has broader enterprise route coverage in the seed, including Anthropic API, AWS Bedrock, Google Vertex AI, Microsoft Foundry, OpenRouter, and Vercel AI Gateway. Gemini 3.5 Flash is tracked through Google AI Studio, Vertex AI, OpenRouter, and Vercel AI Gateway.
Continue comparing
- Claude Sonnet 4.6 vs Gemini 3.5 Pro
- Claude Sonnet 4.6 vs Phi-4 Mini Flash Reasoning
- Claude Opus 4.6 vs Gemini 3.5 Flash
- Claude Sonnet 4.6 vs DeepSeek V4 Flash
- Claude Sonnet 4.6 vs DeepSeek V4 Pro
- Claude Sonnet 4.6 vs Kimi K2.6
- Claude Sonnet 4.6 vs Composer 2.5
- Claude Sonnet 4.6 vs GPT-5.5 Pro
- Claude Sonnet 4.6 vs GLM-5.1
- Claude Opus 4.7 vs Claude Sonnet 4.6
Popular comparisons for Claude Sonnet 4.6
Popular comparisons for Gemini 3.5 Flash
Last reviewed: 2026-06-29. Data sourced from public model cards and provider documentation.