AI model pricing compared: which LLM is cheapest for your codebase?

Choosing an AI model for codebase analysis isn't just about capability — it's about cost. A 200K-token codebase queried 100 times a month adds up fast. This guide breaks down real, verified pricing for the 12 latest models as of mid-2026.

The 12 models we track

Repo2Txt's cost estimator uses pricing sourced directly from official vendor pages. Here's the full list, sorted by input cost (cheapest first):

Model	Context	Input $/M	Output $/M
GPT-5 nano	272K	$0.05	$0.40
Gemini 2.5 Flash-Lite	1M	$0.10	$0.40
GPT-5 mini	922K	$0.25	$2.00
Gemini 2.5 Flash	1M	$0.30	$2.50
DeepSeek V3.2	128K	$0.55	$2.19
Claude Haiku 4.5	200K	$1.00	$5.00
Gemini 2.5 Pro	1M	$1.25	$10.00
GPT-5	400K	$1.25	$10.00
Gemini 3 Pro Preview	1M	$2.00	$8.00
Claude Sonnet 4.5	200K	$3.00	$15.00
Claude Opus 4.5	200K	$5.00	$25.00
GPT-5.5	922K	$5.00	$30.00

Best value by codebase size

Small repos (under 50K tokens)

For small codebases, GPT-5 nano at $0.05/M input is hard to beat. A 50K-token query costs just $0.0025 per call. The trade-off is lower intelligence (rated 3/5), so it's best for simple tasks like generating boilerplate or explaining small functions.

Medium repos (50K–200K tokens)

Gemini 2.5 Flash is the sweet spot. At $0.30/M input with a 1M context window, a 150K-token query costs $0.045. It's fast (5/5 speed) and capable (4/5 intelligence). For most developers, this is the daily driver.

Large repos (200K–500K tokens)

Once you cross 200K tokens, you need models with large context windows. Gemini 2.5 Pro at $1.25/M input handles 1M tokens and offers top-tier reasoning (5/5 intelligence). A 400K-token query costs $0.50 — reasonable for serious codebase analysis.

Huge repos (500K+ tokens)

Only a few models can handle 500K+ token codebases. Gemini 3 Pro Preview ($2.00/M, 1M context) and GPT-5 mini ($0.25/M, 922K context) are your best bets. If your repo exceeds 1M tokens, you'll need to trim it using Repo2Txt's file tree selector.

Round-trip cost: the hidden expense

Input cost is only half the story. When you send 100K tokens to a model, it typically responds with ~30% of that (30K output tokens). The output price is often 4–10x higher than input.

For example, with Claude Opus 4.5:

• Input: 100K tokens × $5.00/M = $0.50
• Output: ~30K tokens × $25.00/M = $0.75
• Total round-trip: $1.25 per call

Repo2Txt's cost estimator shows both input cost and estimated round-trip cost so you can budget accurately.

Conclusion

There's no single "best" model — it depends on your codebase size, task complexity, and budget. Use Repo2Txt's cost estimator to see exact numbers for your specific repo, and pick the model that balances cost and capability for your use case.

For most developers, Gemini 2.5 Flash is the daily driver. Save the expensive models (Claude Opus, GPT-5.5) for complex reasoning tasks where the extra capability justifies the cost.