AI Learning
AI Learning
Which AI coding model to use for each task, what it costs, and how to keep it cheap. Built for a solo developer using Codex and Freebuff.
Last updated: 2026-10-28 · Last checked: 2026-10-28New
Model lineup
Prices are USD per 1M tokens, standard tier. OpenAI prices were verified on OpenAI's pricing page on 28 Oct 2026. Claude and Gemini prices came from secondary sources — check their official pages before budgeting.
| Model | Vendor | Tier | In / 1M | Out / 1M | Cached / 1M | Best for |
|---|---|---|---|---|---|---|
| GPT-6 Astra | OpenAI | Flagship | $10.00 | $50.00 | $1.00 | Hardest reasoning, large refactors, stack-spanning bugsReserve for the hardest 5-10% of tasks. Above 272K input the whole request reprices to $20/$75. |
| GPT-6.1 Sol | OpenAI | Workhorse | $2.00 | $10.00 | $0.100 | Default for most coding workReported Codex default. Near-Astra on agentic coding at a fifth of the price. Cached input is only 5% of input. Above 272K input reprices to $4/$15. |
| GPT-6.1 Terra | OpenAI | Fast / efficient | $1.00 | $5.00 | $0.050 | Docs, writing, UI and screenshot work, fast tasksLaunched 26 Oct 2026. Strong value pick for non-hardest work; Sol/Astra remain better for the most complex coding. |
| GPT-6 Sol | OpenAI | Workhorse | $2.00 | $10.00 | $0.200 | Complex professional and coding workflowsPrevious 6.x workhorse. |
| GPT-6 Luna | OpenAI | Cheap | $0.100 | $0.500 | $0.010 | Quick edits, subagents, file-finding, repeatable tasksCheap enough to be the default subagent model. Above 272K input reprices to $0.20/$0.75. |
| GPT-5.6 Sol | OpenAI | Previous flagship | $4.00 | $20.00 | $0.400 | Previous flagship tierPromo price at least to 21 Nov 2026. |
| GPT-5.6 Terra | OpenAI | Previous mid | $2.00 | $12.00 | $0.200 | Mid-tier workPrevious generation mid tier. |
| GPT-5.6 Luna | OpenAI | Previous efficient | $0.200 | $1.20 | $0.020 | Efficient tasksPrevious generation efficient tier. |
| GPT-5.3 Codex | OpenAI | Codex model | $1.75 | $14.00 | $0.175 | Codex-specific workflowsThe only dedicated -codex model still on the API pricing page. |
| Claude Fable 5.1 | Anthropic | Flagship | $10.00 | $50.00 | $0.250 | Long-horizon reasoning when Opus still failsTry Opus 5.5 first; use Fable only when Opus at higher effort still fails. Flat across the 1M window. |
| Claude Opus 5.5 | Anthropic | Value flagship | $4.00 | $20.00 | $0.200 | Architecture, large refactors, long autonomous runsAnthropic says it performs at Fable 5.1's level on most work at 40% of the price. Cache reads are 5% of input. |
| Claude Sonnet 5.5 | Anthropic | Default | $2.00 | $10.00 | $0.100 | Default for most coding workBest coding-per-dollar in the Claude line on Anthropic's tables. Needs fewer tokens per task than Sonnet 5. |
| Claude Haiku 5.5 | Anthropic | Cheap | $0.100 | $0.500 | $0.010 | Subagents, classification, extraction, bulk editsWatch the 100K cliff: prompts over 100K tokens cost 5x ($0.50/$2.50) for the whole request. |
| Gemini 4 Argon | Flagship | $2.00 | $10.00 | $0.100 | Deep multi-step work, autonomous find-and-patchINTRO PRICE to 30 Nov 2026, then $4/$20. Thinking tokens bill as output, and Argon thinks heavily, so cost per task can exceed the list price. | |
| Gemini 3.1 Pro Preview | Pro | $2.00 | $12.00 | $0.200 | Correctness-critical reasoning, large contextNo free tier. Above 200K tokens: $4/$18, cached $0.40. | |
| Gemini 3.8 Flash | Workhorse | $0.750 | $3.75 | $0.075 | High-volume coding and agents, well-specified featuresINTRO PRICE to 31 Dec 2026, then $1.50/$7.50. Review its output; one independent test scored it well below Pro on correctness. | |
| Gemini 3.5 Flash-Lite | Lite | $0.300 | $2.50 | $0.030 | File-finding, classification, extractionSpecialist subagent tier. | |
| Gemini 3.1 Flash-Lite | Lite | $0.250 | $1.50 | $0.025 | File-finding, research, extractionFreebuff itself uses this tier for file-finding. | |
| Gemini 2.5 Flash-Lite | Cheapest | $0.100 | $0.400 | $0.010 | Cheapest classification and extractionLowest-cost Gemini tier. | |
| DeepSeek V4 Pro | DeepSeek | Budget pro | $1.32 | $3.96 | — | Planning and multi-file reasoning on a budgetPeak price; off-peak $0.66/$1.98. |
| DeepSeek V4.1 Flash | DeepSeek | Budget fast | $0.300 | $1.20 | $0.006 | High-volume work where retries are cheapPeak price; off-peak $0.15/$0.60. Well below Sonnet/Opus on terminal-task benchmarks, so use it where retries are cheap. |
| GLM-5.3 | Zhipu | Open-weight all-rounder | $1.40 | $4.40 | $0.260 | Strongest open-weight all-rounder in this setVendor list price; hoster prices differ. |
| Kimi K2.7 Code | Moonshot | Budget coder | $0.950 | $4.00 | $0.190 | Budget codingVendor list price; hoster prices differ. |
| MiniMax M3 | MiniMax | Budget | $0.300 | $1.20 | — | Budget high-volume workVendor list price; hoster prices differ. |
| Qwen3.8 Max | Alibaba | Budget pro | $2.00 | $6.00 | — | Budget pro-tier workVendor list price; hoster prices differ. |