# Gemini to Bedrock — Model Selection Guide **Applies to:** Vertex AI Generative AI (Gemini models) → Amazon Bedrock This file is loaded by `design-ai.md` when `ai-workload-profile.json` has `summary.ai_source` = `"gemini"` or `"both"`. It provides model mapping tables with pricing and honest competitive analysis for Gemini → Bedrock migration decisions. Verify all pricing against `references/shared/pricing-cache.md`. **Model lifecycle:** Before recommending any Bedrock model, check `references/vendored/ai/ai-model-lifecycle.md`. Do not recommend Legacy models as primary selections for new migrations. Legacy models are annotated below where they appear. **Recommend defaults (Sep 2026):** Claude Sonnet 5 (`anthropic.claude-sonnet-5`) for balanced/flagship; Claude Opus 4.8 for hardest reasoning; Claude Haiku 4.5 for cost/speed. Sonnet 5 is **$2/$10** — the launch rate became the standard price on Sep 1, 2026 (the scheduled increase to $3/$15 was cancelled); comparison tables below use $2/$10. Do not default to any Claude Fable / Mythos frontier model. --- ## Competitive Reality (May 2026) Gemini 3.5 Flash is now GA (May 2026) — the current flagship Flash model. Gemini 3.1 Pro is the current Pro tier. Be honest with users: - Gemini 3.1 Pro leads 13/16 Google-reported benchmarks and 6/10 on the Artificial Analysis Intelligence Index - ARC-AGI-2: 77.1%, SWE-Bench: 80.6% (tied with Opus 4.6 at 80.8%) - Gemini 3.1 Pro costs $2/$12 per 1M tokens — less than half of Opus 4.6 ($5/$25), and slightly above Sonnet 5 ($2/$10, ~13% cheaper blended) - Gemini 3.5 Flash at $1.50/$9.00 is 5x more expensive than the old Gemini 2.5 Flash ($0.30/$2.50) — the Bedrock cost savings case is now much stronger for Flash-tier users **Where Bedrock still wins:** - Claude Sonnet 5 / Opus 4.x lead on real-world agentic tasks (GDPval evaluation) — the gap between benchmarks and production agent reliability is real - Claude prompt caching (90% savings on repeated content) has no Gemini equivalent - Claude function calling remains best-in-class for complex multi-turn tool use - AWS service integration (AgentCore, Knowledge Bases, Guardrails) has no Gemini equivalent **Migration case by tier:** - Gemini 3.5 Flash → Bedrock: **strong cost case** — Nova Lite is 94% cheaper; even Claude Sonnet 5 is close at $2/$10 vs $1.50/$9.00 - Gemini 3.1 Pro → Bedrock: driven by AWS consolidation, agentic reliability, or AWS service breadth; Sonnet 5 at $2/$10 adds a modest (~13%) cost edge, but the case is not cost-led and not general benchmarks - Gemini 3.1 Flash-Lite → Nova Lite/Micro: still 76-88% cheaper, strong cost case - Gemini 2.5 Pro → Bedrock: moderate case (older model) --- ## Bedrock Model Portfolio | Model | Best For | Complexity | Speed | Context | | ---------------- | ------------------------------------------------------------------- | ---------- | ------ | ------- | | Claude Sonnet 5 | Agentic tasks, tool use | High | High | 1M | | Claude Opus 4.6 | Maximum reasoning | High | Medium | 200K | | Claude Haiku 4.5 | Simple + fast | Medium | High | 200K | | Llama 4 Maverick | Cost-effective + multimodal | Medium | High | 1M | | Llama 4 Scout | Ultra-long context, cheapest | Medium | Medium | 10M | | Nova 2 Pro | AWS flagship, multimodal | High | High | 1M | | Nova 2 Lite | AWS mid-tier, long context | Medium | High | 1M | | Nova Pro | AWS balanced | Medium | High | 300K | | Nova Lite | AWS fast + cheapest | Medium | High | 300K | | Nova Micro | AWS fastest, text-only | Low | High | 128K | | Nova Premier | Complex reasoning — **Legacy** (EOL Sep 14, 2026) prefer Nova 2 Pro | High | Medium | 1M | | DeepSeek-R1 | Chain-of-thought reasoning | High | Medium | 128K | | Mistral Large 3 | EU/Multilingual | High | Medium | 256K | --- ## Gemini → Bedrock Model Mapping ### Gemini Pro Tier | Gemini Model | Price (in/out per 1M) | Best Bedrock Match | Bedrock Price | Winner | | -------------- | --------------------- | ------------------ | -------------- | ------------------- | | Gemini 3.1 Pro | $2.00 / $12.00 | Claude Sonnet 5 | $2.00 / $10.00 | Bedrock 13% cheaper | | Gemini 3.1 Pro | $2.00 / $12.00 | Claude Opus 4.8 | $5.00 / $25.00 | Gemini 54% cheaper | | Gemini 3.1 Pro | $2.00 / $12.00 | Nova 2 Pro | $1.38 / $11.00 | Bedrock 14% cheaper | | Gemini 3 Pro | $0.50 / $3.00 | Llama 4 Maverick | $0.24 / $0.97 | Bedrock 64% cheaper | | Gemini 3 Pro | $0.50 / $3.00 | Llama 4 Scout | $0.17 / $0.66 | Bedrock 75% cheaper | | Gemini 3 Pro | $0.50 / $3.00 | Nova Pro | $0.80 / $3.20 | Gemini 17% cheaper | | Gemini 2.5 Pro | $1.25 / $10.00 | Claude Sonnet 5 | $2.00 / $10.00 | Gemini 11% cheaper | | Gemini 2.5 Pro | $1.25 / $10.00 | Nova Pro | $0.80 / $3.20 | Bedrock 62% cheaper | | Gemini 2.5 Pro | $1.25 / $10.00 | Nova 2 Pro | $1.38 / $11.00 | Gemini 9% cheaper | > **Gemini 3.1 Pro breakpoint:** $4.00/$18.00 per 1M for prompts >200k tokens. Table above uses ≤200k rates. ### Gemini Flash/Lite Tier | Gemini Model | Price (in/out per 1M) | Best Bedrock Match | Bedrock Price | Winner | | ------------------------- | ----------------------------------------------------------- | -------------------------------------- | -------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Gemini 3.5 Flash (GA) | $1.50 / $9.00 | Nova Lite | $0.06 / $0.24 | Bedrock 94% cheaper — strong migration case; 3.5 Flash is 5x more expensive than old 2.5 Flash | | Gemini 3.5 Flash (GA) | $1.50 / $9.00 | Claude Sonnet 5 | $2.00 / $10.00 | Gemini 14% cheaper — but Sonnet leads on agentic reliability | | Gemini 3.1 Flash-Lite | $0.25 / $1.50 | Nova Lite | $0.06 / $0.24 | Bedrock 76% cheaper | | Gemini 3.1 Flash-Lite | $0.25 / $1.50 | Nova Micro | $0.035 / $0.14 | Bedrock 88% cheaper | | Gemini 2.5 Flash | $0.30 / $2.50 | Nova Lite | $0.06 / $0.24 | Bedrock 88% cheaper | | Gemini 2.5 Flash Thinking | $0.30 / $0.60–$3.50 (varies by thinking budget) | Claude Sonnet 5 with extended thinking | $2.00 / $10.00 | Gemini is cheaper at the listed rates: even at $3.50/M output, its 2:1 blended rate is ~$1.37/M vs ~$4.67/M for Sonnet 5 (~71% lower). Consider Sonnet 5 for capability requirements, not expected savings; profile each model's billed thinking tokens. | | Gemini 2.0 Flash | $0.10 / $0.40 | Nova Micro | $0.035 / $0.14 | Bedrock 65% cheaper | | Gemini Flash 1.5 | **Legacy — EOL Sep 24, 2025. Migrate to Gemini 3.5 Flash.** | Nova Lite | $0.06 / $0.24 | If still in use, migrate source model first; strong Bedrock cost case once on 3.5 Flash | ### Legacy/Specialized | Gemini Model | Price (in/out per 1M) | Best Bedrock Match | Bedrock Price | Winner | | ------------------------------------------------------------ | ----------------------------------------------------------------------------- | --------------------------------- | ------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------- | | Gemini 1.5 Pro | **Legacy — EOL Sep 24, 2025. Migrate to Gemini 2.5 Pro or 3.x Pro.** | Claude Sonnet 5 | $2.00 / $10.00 | If still in use, migrate source model first | | text-bison / chat-bison | Legacy | Llama 4 Scout | $0.17 / $0.66 | Bedrock (better quality + cheaper) | | text-embedding-004 | $0.025 / N/A | Titan Embeddings V2 | $0.02 / N/A | Bedrock 20% cheaper | | Gemini Embedding 2 (multimodal — text/image/video/audio/PDF) | text $0.20/1M, image $0.45/1M ($0.00012/img), audio $6.50/1M, video $12.00/1M | Amazon Nova Multimodal Embeddings | text $0.135/1M, image $0.00006/unit (see `shared/pricing-cache.md` § Multimodal embeddings) | Bedrock ~32% cheaper on text, ~50% cheaper on image | | imagen-* | Varies | Stable Image Core | $0.04/img | Nova Canvas excluded (EOL Sep 30, 2026); Ultra $0.08/img if quality-first | _Percentages are blended savings using a 2:1 input-to-output token ratio. Actual savings depend on your input/output ratio._ --- ## Decision Paths by Priority ### Quality-First Gemini 3.1 Pro Preview matches or beats Opus 4.6 on most reasoning benchmarks at less than half the cost. Be transparent: - If user needs **general reasoning/coding quality** → Gemini 3.1 Pro is competitive or better. Migration case is weak unless driven by AWS consolidation. - If user needs **agentic reliability** (real-world multi-step tasks) → **Claude Sonnet 5** still leads on GDPval. This is the honest differentiator. - If user needs **maximum reasoning on hardest problems** → **Claude Opus 4.8** ($5/$25 headline on-demand, same tier as Opus 4.6) — use the latest [Claude on Bedrock](https://aws.amazon.com/bedrock/pricing/) model card for benchmark deltas vs Gemini; Opus 4.6 remains a same-price alternative where batch or regional availability matters. ### Speed-First Gemini Flash → **Nova Micro** (<200ms, text-only, cheapest), **Haiku 4.5** (<400ms, vision), or **Llama 4 Scout** (<300ms, cheapest capable) ### Cost-First - Gemini Flash/Lite → **Nova Lite** (54-88% cheaper), **Nova Micro** (53-64% cheaper) - Gemini Pro → **Llama 4 Maverick** ($0.24/$0.97, 63% cheaper than Gemini 3 Pro) or **Llama 4 Scout** ($0.17/$0.66, 75% cheaper) ### Balanced - Gemini 3.1 Pro → **Nova 2 Pro** (-14% cost, AWS-native) or **Claude Sonnet 5** (-13% cost, stronger agentic reliability) - Gemini 2.5 Pro → **Nova 2 Pro** (+10% cost, AWS-native) or **Nova Pro** (-62% cost) - Gemini 3 Pro → **Llama 4 Maverick** (-63%), **Nova Pro** (+20%) --- ## Volume-Based Recommendations **Low (<1M tokens/day):** Use best model for quality. Cost difference minimal at this volume. **Medium (1-10M tokens/day):** Present cost comparison at volume. At 5M input + 2.5M output/day: | Model | Monthly Cost | | ---------------- | -------------- | | Gemini 3 Pro | $300 | | Llama 4 Maverick | $109 (-64%) | | Llama 4 Scout | $75 (-75%) | | Nova Pro | $360 (+20%) | | Claude Sonnet 5 | $1,050 (+250%) | **High (10-100M tokens/day):** Cost optimization critical. Recommend multi-model tiered approach. Llama 4 Maverick/Scout or Nova for output-heavy workloads. **Very high (>100M tokens/day):** Mandatory multi-model tiered strategy: - Simple tasks (60% of traffic) → Nova Micro or Llama 4 Scout - Moderate tasks (30% of traffic) → Llama 4 Maverick or Nova Pro - Complex tasks (10% of traffic) → Claude Sonnet 5 --- ## Cost Comparison Table (150M input + 75M output per month) | Gemini Model | Monthly | Best Bedrock Match | Monthly | Difference | | ------------------------------- | ------- | ------------------------------ | ------- | ---------- | | Gemini 3.1 Pro Preview ($2/$12) | $1,200 | Claude Sonnet 5 ($2/$10) | $1,050 | -13% | | Gemini 3.1 Pro Preview ($2/$12) | $1,200 | Claude Opus 4.8 ($5/$25) | $2,625 | +54% | | Gemini 3.1 Pro Preview ($2/$12) | $1,200 | Nova 2 Pro ($1.38/$11.00) | $1,032 | -14% | | Gemini 3 Pro ($0.50/$3.00) | $300 | Llama 4 Maverick ($0.24/$0.97) | $109 | -64% | | Gemini 3 Pro ($0.50/$3.00) | $300 | Llama 4 Scout ($0.17/$0.66) | $75 | -75% | | Gemini 2.5 Pro ($1.25/$10) | $938 | Nova 2 Pro ($1.38/$11.00) | $1,032 | +9% | | Gemini 2.5 Pro ($1.25/$10) | $938 | Nova Pro ($0.80/$3.20) | $360 | -62% | | Gemini 2.5 Flash ($0.30/$2.50) | $233 | Nova Lite ($0.06/$0.24) | $27 | -88% | | Gemini 2.0 Flash ($0.10/$0.40) | $45 | Nova Micro ($0.035/$0.14) | $16 | -64% | _Difference column shows blended savings at a 2:1 input/output token ratio. Positive = Bedrock costs more (Gemini cheaper), negative = Bedrock cheaper._ --- ## Prompt Caching (Claude Only) Cache frequently-used system prompts for 90% cost reduction on cached portions. Example: 10K token system prompt repeated 1000x → $30 without caching, $3 with caching. Not available on other Bedrock models. This is a significant Claude advantage for applications with heavy system prompt repetition. --- ## Feature Migration Notes | Gemini Feature | Bedrock Equivalent | Notes | | ------------------------------------------------ | ------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Function calling | Claude tools (excellent), Mistral (good) | Minimal changes | | Structured output/JSON | Claude (excellent), Nova Pro (good) | Most models via prompt | | Streaming | All major models | Same SSE pattern | | Vision | Claude Sonnet/Haiku, Llama 4 Maverick | Multimodal parity | | Context caching | Claude prompt caching | Different mechanics, not a drop-in: Gemini uses explicit TTL-based `cachedContent` objects you create/reference by name; Claude uses inline cache-control breakpoints on the request itself. Re-implement the caching call sites, don't just swap endpoints. 90% savings on cached portions once re-implemented. | | Audio/video input | Nova 2 Sonic (speech), Transcribe/Rekognition (preprocessing) | Nova Sonic v1 is Legacy; use Nova 2 Sonic | | Embeddings (text-only) | Amazon Titan Embeddings ($0.02/1M, 1536 dims) | Must re-embed all docs | | Embeddings (multimodal — text+image/video/audio) | Amazon Nova Multimodal Embeddings (see pricing table above) | Do NOT route to the text-only Titan Embeddings row above — Gemini Embedding 2 unifies text+image+video+audio into one model, so detect by call arguments (image/blob content present), not by model name alone. Must re-embed all docs into the new vector space regardless of target. |