Workload Benchmark & Cost Model
Code Generation & AI IDE Agent Cost Comparison
Repository-aware code generation sends a large cacheable project context—repo map, architecture rules, and tool definitions (12,000 tokens)—plus active file diffs (2,500 dynamic tokens) and generates long code + reasoning output (1,600 tokens) at 500 requests/day.
Code Generation Cost Comparison (20 models)
100% Client-Side · Zero Prompt TransmissionAvg input/turn: 14,500 tok · Daily turns: 500 · Registry verified: 2026-11-12
| Model & Host | Active Tier & Rules | Eff. In / Out ($/1M) | Cache Read ($/1M) | Daily | Monthly (30d) | $/1K Turns | Savings vs Sync |
|---|---|---|---|---|---|---|---|
LOWESTGPT-6 Luna OpenAI·OpenAI First-Party·922K max in Verified 2026-11-12Official Docs ↗ | Cache On Standard (<= 272K) | $0.100 $0.500 | $0.0100 | $0.71 | $21.37 | $1.424 | -36.7% $12/mo saved |
Anthropic·Anthropic First-Party·1000K max in Verified 2026-11-12Official Docs ↗ | Cache On Standard (<= 100K) | $0.100 $0.500 | $0.0100 | $0.71 | $21.37 | $1.424 | -36.7% $12/mo saved |
Mistral Small 4 (2603)Open-Weight Mistral·Mistral La Plateforme·256K max in Verified 2026-11-12Official Docs ↗ | Cache On Standard (<= 256K) | $0.150 $0.600 | $0.0150 | $0.92 | $27.58 | $1.839 | -41.3% $19/mo saved |
GPT-OSS 120B (Groq)Open-Weight Groq·Groq LPU·131K max in Verified 2026-11-12Official Docs ↗ | Cache On Standard (<= 131K) | $0.150 $0.600 | $0.0750 | $1.21 | $36.23 | $2.415 | -23.0% $11/mo saved |
GPT-OSS 120B (Together AI)Open-Weight Together AI·Together AI Serverless (MXFP4)·131K max in Verified 2026-11-12Official Docs ↗ | Standard (<= 131K)
| $0.150 $0.600 | — | $1.57 | $47.03 | $3.135 | 0.0% |
GPT-OSS 120B (Amazon Bedrock)Open-Weight AWS Bedrock·Amazon Bedrock (US Regions)·128K max in Verified 2026-11-12Official Docs ↗ | US Standard (<= 128K)
| $0.150 $0.600 | — | $1.57 | $47.03 | $3.135 | 0.0% |
DeepSeek Flash (V4.1-Flash-1017)Open-Weight DeepSeek·DeepSeek First-Party·1000K max in Verified 2026-11-12Official Docs ↗ | Cache On Standard (<= 1M) | $0.300 $1.200 | $0.0060 | $1.72 | $51.71 | $3.448 | -45.0% $42/mo saved |
DeepSeek V4.1 Flash (Together AI)Open-Weight Together AI·Together AI Serverless (FP8)·1000K max in Verified 2026-11-12Official Docs ↗ | Cache On Standard (<= 1M) | $0.300 $1.200 | $0.0060 | $1.72 | $51.71 | $3.448 | -45.0% $42/mo saved |
Llama 4 Maverick 17B (Amazon Bedrock)Open-Weight AWS Bedrock·Amazon Bedrock (US Regions)·1000K max in Verified 2026-11-12Official Docs ↗ | US Standard (<= 1M)
| $0.240 $0.970 | — | $2.52 | $75.48 | $5.032 | 0.0% |
Mistral Large 4Open-Weight Mistral·Mistral La Plateforme·1000K max in Verified 2026-11-12Official Docs ↗ | Cache On Standard (<= 1M) | $0.700 $2.100 | $0.0700 | $3.73 | $111.93 | $7.462 | -44.8% $91/mo saved |
Google·Google AI Studio / Gemini API·1049K max in Verified 2026-11-12Official Docs ↗ | Cache OnPromo ends 2026-12-31 Intro Promo (<= 1M, thru 2026-12-31)
| $0.750 $3.750 | $0.0750 | $5.29 | $158.81 | $10.587 | -37.3% $94/mo saved |
DeepSeek V4 Pro (0813)Open-Weight DeepSeek·DeepSeek First-Party·1000K max in Verified 2026-11-12Official Docs ↗ | Cache On Standard (<= 1M) | $1.320 $3.960 | $0.0440 | $6.61 | $198.40 | $13.226 | -48.1% $184/mo saved |
Qwen3.8 27B Preview (Groq)Open-Weight Groq·Groq LPU·131K max in Verified 2026-11-12Official Docs ↗ | Standard (<= 131K)
| $0.800 $4.000 | — | $9.00 | $270.00 | $18.000 | 0.0% |
Google·Google AI Studio / Gemini API·1049K max in Verified 2026-11-12Official Docs ↗ | Cache OnPromo ends 2026-11-30 Intro Promo (<= 1M, thru 2026-11-30)
| $2.000 $10.000 | $0.1000 | $13.38 | $401.40 | $26.760 | -40.5% $274/mo saved |
OpenAI·OpenAI First-Party·922K max in Verified 2026-11-12Official Docs ↗ | Cache On Standard (<= 272K) | $2.000 $10.000 | $0.1000 | $13.76 | $412.92 | $27.528 | -38.8% $262/mo saved |
Anthropic·Anthropic First-Party·1000K max in Verified 2026-11-12Official Docs ↗ | Cache On Standard (<= 1M) | $2.000 $10.000 | $0.1000 | $13.76 | $412.92 | $27.528 | -38.8% $262/mo saved |
Claude Sonnet 5.5 (Amazon Bedrock) AWS Bedrock·Amazon Bedrock (Cross-Region Inference)·1000K max in Verified 2026-11-12Official Docs ↗ | Cache On Global CRIS (<= 1M) | $2.000 $10.000 | $0.1000 | $13.76 | $412.92 | $27.528 | -38.8% $262/mo saved |
Gemini 3.1 Pro Preview Google·Google AI Studio / Gemini API·1049K max in Verified 2026-11-12Official Docs ↗ | Cache On Standard (<= 200K) | $2.000 $12.000 | $0.2000 | $16.32 | $489.72 | $32.648 | -32.3% $233/mo saved |
Claude Fable 5.2 Anthropic·Anthropic First-Party·1000K max in Verified 2026-11-12Official Docs ↗ | Cache On Standard (<= 1M) | $8.000 $40.000 | $0.2000 | $54.10 | $1,622.88 | $108.192 | -39.9% $1,077/mo saved |
OpenAI·OpenAI First-Party·922K max in Verified 2026-11-12Official Docs ↗ | Cache On Standard (<= 272K) | $10.000 $50.000 | $1.0000 | $71.22 | $2,136.60 | $142.440 | -36.7% $1,238/mo saved |
Budgeting & Tokenizer Disclaimer: Estimates assume USD list pricing verified on 2026-11-12. Actual token counts vary across provider tokenizers (for example, Claude 4.7+ uses a newer tokenizer that generates ~30% more tokens on identical prose than legacy models), and reasoning models bill hidden thinking tokens as output tokens.
Token Anatomy & Scaling Assumptions
- Cacheable Prefix (12,000 tokens): Repository tree, coding standards, type definitions, and MCP/tool schemas shared across edits.
- Dynamic Input (2,500 tokens): Currently open files, compiler errors, and user edit instructions.
- Completion Output (1,600 tokens): Generated code patches, tests, and reasoning tokens (billed at output rate, which is typically 3x–5x input rate).
- Daily Scale: 500 requests/day over 16 active hours/day (15,000 requests/month).
Worked Note: At 12,000 prefix / 2,500 dynamic / 1,600 output and 500 req/day (80% hit, 4 writes/hr over 16h), Claude Sonnet 5.5 ($2.00/$10.00, $0.10 read, $2.50 5m write) costs $13.76/day ($412.92/mo) compared to $675.00/mo uncached—saving $262.08/mo (38.8%). Notice that output tokens ($240.00/mo on Sonnet 5.5) still account for 58.1% of the cached bill.
Caching & Batch Strategy
Because the 12,000-token repo prefix makes up 82.8% of total input tokens and exceeds every provider's minimum cache threshold (including Gemini's 4,096 tokens), prompt caching is essential. With 80% cache hits and 4 cold writes/hour (64 writes/day), providers with 95%–98% cache-read discounts (GPT-6.1 Sol, Claude Sonnet 5.5, DeepSeek V4.1 Flash) dramatically slash input spend.
Interactive IDE edits cannot wait for 24-hour Batch queues, so Batch API is off by default. Enable Batch only for asynchronous overnight PR test generation, repo-wide lint/migration sweeps, or CI benchmark evaluations.