GPT-6.1 Sol Is Live: Near-Astra Smarts, Half-Price Cache Reads

​

Key Takeaways

• Cache reads cut in half: gpt-6.1-sol keeps the same $2 input / $10 output as gpt-6-sol, while cache reads drop from $0.20 to $0.10 per 1M tokens, just 5％ of the standard input price. Cache-heavy workloads such as multi-turn chat, agents and coding assistants will see a noticeably smaller bill
• Close to Astra: 52 on the Artificial Analysis Intelligence Index at max effort, 4 points above GPT-6 Sol and only 1 point below GPT-6 Astra. OpenAI says it matches Astra on DeepSWE v1.1 at roughly one-fifth the price
• Fewer factual errors: on deliberately difficult factuality prompts at low reasoning effort, OpenAI reports the share of responses containing a factual error falling from 11.4％ to 7.7％
• Live on APIYI: all four billing items match the provider with no markup; savings come from our top-up promotions, see Top-up Promotions
• One change to note: reasoning effort no longer supports none; the lowest level is low

​

Background

​

In Detail

​

Benchmarks

​

Key Features

​

Specifications

​

Practical Use

​

Recommended Scenarios

• Coding assistants and agents: system prompts and code context are re-read from cache every turn, so halved cache reads show up directly on the bill
• Long-document analysis: ask repeated questions about the same contract, financial report or codebase, with cached tokens billed at just $0.10
• Mid-difficulty tasks currently on Astra: run a comparison with GPT-6.1 Sol first; capability is close and the price is about one-fifth

​

Code Examples

from openai import OpenAI

client = OpenAI(api_key="sk-your-apiyi-key", base_url="https://api.apiyi.com/v1")

response = client.responses.create(
    model="gpt-6.1-sol",
    reasoning={"effort": "medium"},   # low / medium / high / xhigh / max
    input="Review this code, point out potential bugs and suggest fixes: ...",
)
print(response.output_text)

from openai import OpenAI

client = OpenAI(api_key="sk-your-apiyi-key", base_url="https://api.apiyi.com/v1")

response = client.chat.completions.create(
    model="gpt-6.1-sol",
    reasoning_effort="medium",
    messages=[
        {"role": "user", "content": "Summarize the core goals of this requirements doc in three sentences: ..."}
    ]
)
print(response.choices[0].message.content)

curl https://api.apiyi.com/v1/responses \
  -H "Authorization: Bearer sk-your-apiyi-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6.1-sol",
    "reasoning": {"effort": "medium"},
    "input": "In one sentence, what changed from GPT-6 Sol to GPT-6.1 Sol?"
  }'

​

Best Practices

• Migrating from GPT-6 Sol is mostly a model-name change: gpt-6-sol → gpt-6.1-sol. The one thing to check is reasoning effort: code that passes none must switch to low
• Prefer Responses for tool calling: OpenAI’s model docs list tool calling under the Responses API, so use /v1/responses for function calling and multi-step agents
• Put stable content first: place system prompts, tool definitions and long documents at the start of the message so later turns hit the cache at $0.10
• Stay within 272K input where possible: above that, the whole request moves to a higher pricing tier (see the tier table below)
• Leave room for reasoning: reasoning tokens count toward the output limit, so do not set max_output_tokens too tight

​

Pricing and Availability

​

Pricing

​

Long-Context Tiers

​

What Half-Price Cache Reads Save

​

Groups and Endpoints

​

Stack With Top-up Promotions

​

Summary and Recommendations

1. Current gpt-6-sol users: switch to gpt-6.1-sol and check whether you pass reasoning_effort: none; nothing else needs to change
2. Using gpt-6-astra for mid-difficulty tasks: run your own tasks through both; if capability is close, cost drops to about one-fifth
3. Cache-heavy workloads: keep a stable prefix at the very start of your messages to get the most out of the halved cache price