Cost to code a small feature with AI
Scenario: one request, 1,500 words of instruction and existing code in, 750 words of code and notes out. No agent loop.
Sorted by the calculated USD amount, lowest first. A sponsored label would not move a row or change a figure. How the figures are made.
400 small features a month: AI about A$0.40 cheapest, about A$8.03 typical. By hand about 800 hours, about A$23,560.00.400 small features a month: AI about US$0.28 cheapest, about US$5.60 typical. By hand about 800 hours, about US$16,433.10.
Starting wage is the cited award minimum, not a loaded cost. Fair Work Ombudsman, Clerks—Private Sector Award 2020. A USD figure here is that AUD wage converted with the same RBA rate as the AI prices.
- =1
GPT-6 Luna
Source · read 20 Oct 2026
A$1.00US$0.70
per 1,000 small featuresper 1,000 small features
Assumptions
Scenario: one request, 1,500 words of instruction and existing code in, 750 words of code and notes out. No agent loop. Input 1500 words counted as 2000 tokens. Output 750 words counted as 1000 tokens. OpenAI's token help article says that for English, 100 tokens are approximately 75 words. The same count is used for every vendor. It is not that vendor's tokenizer. Not included: reasoning tokens, tool calls, retries, cache writes, and any second request. A vendor bill can be higher than this row. Standard short-context list price on the OpenAI pricing page: US$0.10 input and US$0.50 output per 1M tokens. That page defines short context as 272,000 input tokens or fewer. No cached-input discount is applied.
Formula: (2000 / 1,000,000) × US$0.10 + (1000 / 1,000,000) × US$0.50.
Pricing basis
Per-token API list price (standard, short context)
Australian business notes
OpenAI's data-controls guide lists Australia at au.api.openai.com with regional storage yes and regional processing no. It says non-US regions require abuse-monitoring approval and a modified-retention amendment, and that without regional processing OpenAI may process and temporarily store customer content outside the region. An asterisk on Australia says image support needs approval for enhanced zero data retention or enhanced modified abuse monitoring. The API pricing page says data-residency endpoints cost 10% more for eligible models released on or after 5 March 2026 (https://developers.openai.com/api/docs/pricing). The USD figure in this row is the standard list price, not that uplift. The pricing page does not mention AUD or GST. AUD in this row is USD divided by 0.6975, the RBA United States dollar rate per Australian dollar on 2026-10-19. Australian note source.
Exact figure
Calculated total: US$0.0007 (A$0.001004). These are the unrounded totals. The figure above is a shorter reading of the same amounts.
- =1
Claude Haiku 5.5
Source · read 20 Oct 2026
A$1.00US$0.70
per 1,000 small featuresper 1,000 small features
Assumptions
Scenario: one request, 1,500 words of instruction and existing code in, 750 words of code and notes out. No agent loop. Input 1500 words counted as 2000 tokens. Output 750 words counted as 1000 tokens. OpenAI's token help article says that for English, 100 tokens are approximately 75 words. The same count is used for every vendor. It is not that vendor's tokenizer. Not included: reasoning tokens, tool calls, retries, cache writes, and any second request. A vendor bill can be higher than this row. Claude Haiku 5.5 list price for a prompt up to 100,000 tokens: US$0.10 input and US$0.50 output per 1M tokens. This scenario is under that prompt length. Global routing price, not the US-only 1.1× price. No cache discount.
Formula: (2000 / 1,000,000) × US$0.10 + (1000 / 1,000,000) × US$0.50.
Pricing basis
Per-token API list price (prompt up to 100,000 tokens)
Australian business notes
Claude's data-residency page lists inference geo as global or us, and says the only workspace geo is us. It does not list Australia. The pricing page says US-only inference on Claude 4.6 and later is 1.1 times the token price, and that global routing uses the standard price (https://platform.claude.com/docs/en/about-claude/pricing). This row uses the standard price. That pricing page says all prices are in USD and does not mention GST. AUD in this row is USD divided by 0.6975, the RBA United States dollar rate per Australian dollar on 2026-10-19. Australian note source.
Exact figure
Calculated total: US$0.0007 (A$0.001004). These are the unrounded totals. The figure above is a shorter reading of the same amounts.
- =3
Claude Sonnet 5.5
Source · read 20 Oct 2026
A$20.07US$14.00
per 1,000 small featuresper 1,000 small features
Assumptions
Scenario: one request, 1,500 words of instruction and existing code in, 750 words of code and notes out. No agent loop. Input 1500 words counted as 2000 tokens. Output 750 words counted as 1000 tokens. OpenAI's token help article says that for English, 100 tokens are approximately 75 words. The same count is used for every vendor. It is not that vendor's tokenizer. Not included: reasoning tokens, tool calls, retries, cache writes, and any second request. A vendor bill can be higher than this row. Claude Sonnet 5.5 list price: US$2 input and US$10 output per 1M tokens. Global routing price, not the US-only 1.1× price. No cache discount.
Formula: (2000 / 1,000,000) × US$2 + (1000 / 1,000,000) × US$10.
Pricing basis
Per-token API list price (standard global routing)
Australian business notes
Claude's data-residency page lists inference geo as global or us, and says the only workspace geo is us. It does not list Australia. The pricing page says US-only inference on Claude 4.6 and later is 1.1 times the token price, and that global routing uses the standard price (https://platform.claude.com/docs/en/about-claude/pricing). This row uses the standard price. That pricing page says all prices are in USD and does not mention GST. AUD in this row is USD divided by 0.6975, the RBA United States dollar rate per Australian dollar on 2026-10-19. Australian note source.
Exact figure
Calculated total: US$0.014 (A$0.020072). These are the unrounded totals. The figure above is a shorter reading of the same amounts.
- =3
Gemini 4 Argon
Source · read 20 Oct 2026
A$20.07US$14.00
per 1,000 small featuresper 1,000 small features
Assumptions
Scenario: one request, 1,500 words of instruction and existing code in, 750 words of code and notes out. No agent loop. Input 1500 words counted as 2000 tokens. Output 750 words counted as 1000 tokens. OpenAI's token help article says that for English, 100 tokens are approximately 75 words. The same count is used for every vendor. It is not that vendor's tokenizer. Not included: reasoning tokens, tool calls, retries, cache writes, and any second request. A vendor bill can be higher than this row. Gemini 4 Argon standard paid-tier list price as printed for 20 Oct 2026: US$2.00 input and US$10.00 output per 1M tokens through 30 Nov 2026, then US$4.00 and US$20.00 from 1 Dec 2026. This row uses the through-30-Nov figures. The page says the output price includes thinking tokens. This row still multiplies only the scenario's visible-text token count, because a thinking-token count for the scenario is not published.
Formula: (2000 / 1,000,000) × US$2.00 + (1000 / 1,000,000) × US$10.00.
Pricing basis
Per-token API list price (standard paid tier, introductory rate)
Australian business notes
The Gemini API available-regions page lists Australia among countries where the Gemini API and Google AI Studio are available. It does not describe an Australian storage or processing region. The pricing page labels paid prices as USD and does not mention GST. The AUD figure is this site's conversion, not a price Google shows in AUD. AUD in this row is USD divided by 0.6975, the RBA United States dollar rate per Australian dollar on 2026-10-19. Australian note source.
Exact figure
Calculated total: US$0.014 (A$0.020072). These are the unrounded totals. The figure above is a shorter reading of the same amounts.
Get the weekly AI cost update
Free. One email a week. Unsubscribe anytime.