GPT 5.6 Sol 20% price reduction
TL;DR Highlight
GPT-5.6 Sol drops to $4 input and $20 output, cheaper than GPT-5.5
Who Should Read
A backend developer comparing GPT-5.5 or GPT-5.4 API costs and deciding whether to switch models
Core Mechanics
- Sol costs $4 input, $20 output
- Prices are per 1M tokens, with cached input at $0.40.
- The doc calls it a 20% input and 33% output reduction.
- GPT-5.5 input is $5, GPT-5.4 is $2.50
- These are the input prices in the doc's comparison table.
- Sol is $1 cheaper than GPT-5.5 and $1.50 above GPT-5.4 on input.
- Promo pricing runs through at least Nov 21, 2026
- The doc says the promotional price is available until at least that date.
- It gives no price for after that.
- Over 272K input tokens, the whole request costs more
- Input is priced at 2x and output at 1.5x.
- Cache writes are billed at 1.25x the uncached input rate.
- 1.05M context, 128K max output
- Knowledge cutoff is Feb 16, 2026.
Community Reactions
- The cut is only for 3 months, commenters said
- Some asked for the headline to be corrected.
- Others thought it might become permanent.
- chat-latest pricier than Sol annoyed users
- One commenter said non-thinking instant-model users would pick chat-latest.
- They found it absurd that it costs more than Sol.
- GitHub Copilot showed Sol at 2/10, one user reported
- It was listed below Terra, so they wondered if it was an error.
- Competition is seen as the driver
- Several commenters credited low-cost rivals like DeepSeek.
- One asked whether subscription usage limits will rise too.
How to Apply
- Recompute your monthly cost with the new prices
- Pull last month's input, output, and cached token counts.
- Multiply by $4, $0.40, and $20 and compare with your current model.
- Check the share of requests over 272K input tokens
- Count requests above 272K in your token logs.
- Reprice those at 2x input and 1.5x output.
- Put the promo end date in your cost assumptions
- Assume the price may change after Nov 21 and build a buffer scenario.
- Decide whether to pin a snapshot instead of the alias.
- Compare cost across reasoning.effort levels
- Run the same eval set at medium and high and log output tokens.
- Output costs 5x input, so check output growth first.
Try It
# Paste into Claude or ChatGPT. Prompt for comparing monthly API cost
Calculate my service's monthly cost with the price table below. Prices are per 1M tokens.
- GPT-5.6 Sol: input $4, cached input $0.40, output $20
- GPT-5.5: input $5 (I will give cached and output prices below)
- GPT-5.4: input $2.50 (I will give cached and output prices below)
- For requests over 272K input tokens, apply 2x input and 1.5x output to the whole request.
- Cache writes are billed at 1.25x the uncached input rate.
My monthly usage:
- Regular input tokens: {{monthly input tokens}}
- Cache read tokens: {{monthly cached tokens}}
- Output tokens: {{monthly output tokens}}
- Share of requests over 272K: {{ratio}}
Show monthly cost per model in a table with the formulas. Do not guess missing prices; ask me.
Next step: ask for a scenario where the promo ends and the price goes up.Terminology
Related Papers
Mercury 3: Diffusion at scale
Scaling diffusion LLM 40x gives Mercury 3: 4–8x faster than speed-tier models, similar quality
Timestral: Mistral's time-series foundation models
A 209M time-series model ranks first on three benchmarks and runs on CPU.
DeepSeek V4.1 Flash 1017 (Max)
DeepSeek V4.1 Flash 1017 Max scores 45 on the Intelligence Index at 211 tok/s and $0.30 input.
If you don't run your own LLM proxy, how do you know whats going on?
Running your own LLM proxy shows which keys, models, and cache misses drive your usage.
DeepSeek: DeepSeek v4.1 Flash 1017
DeepSeek V4.1 Flash 1017: $0.05/M input, 1M context, built for agents
DeepSeek-v4.1-Flash-1017
DeepSeek V4.1 Flash minor update lifts coding and agent benchmarks while cutting run time and output length
Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardware