ModelPort

Migrating to GPT-6 (Astra, 6.1 Sol, Sol, Luna): breaking changes and fixes

What changes when you move code from GPT-5.x to the GPT-6 family, the fix for each, and what to re-tune.

Model IDs: gpt-6-astra, gpt-6.1-sol, gpt-6-sol, gpt-6-luna Context / output: 1,050,000 tokens (922,000 max input) / 128,000 tokens Knowledge cutoff: Apr 30, 2026 (Astra and 6.1 Sol) Last verified: 2026-10-11 Source: OpenAI “Using GPT-6” guide and the gpt-6-astra and gpt-6.1-sol model pages. Sol and Luna prices and all release dates come from secondary sources and are marked below. Re-check the OpenAI model pages before quoting.

Variants

Variant	Model ID	Use case	Price in / out per 1M tokens
GPT-6 Astra	gpt-6-astra	Highest intelligence: demanding reasoning, coding, professional work	$10 / $50 (cached $1, cache writes $12.50)
GPT-6.1 Sol	gpt-6.1-sol	Balanced speed, cost, intelligence	$2 / $10 (cached $0.10, cache writes $2.50)
GPT-6 Sol	gpt-6-sol	Existing model. Review the migration notes before moving to 6.1 Sol	$2 / $10 (secondary source)
GPT-6 Luna	gpt-6-luna	Fastest, cheapest, high-volume focused tasks	$0.10 / $0.50 (secondary source)

Prompts over 272K input tokens bill the whole request at 2x input and cache rates and 1.5x output. Fast mode is 2x, Ultrafast is 6x, Batch and Flex are 50% lower, regional processing adds 10% (6.1 Sol page).

Secondary-source release dates (unverified): Astra 2026-09-03, GPT-6 Sol and Luna 2026-09-22, GPT-6.1 Sol 2026-09-29 or 2026-09-30.

Reasoning effort

Set reasoning.effort (Responses API) or reasoning_effort (Chat Completions).

Model	Levels	Notes
GPT-6.1 Sol	low, medium (default), high, xhigh, max	none and minimal unsupported
GPT-6 Astra	low, medium, high, xhigh, max	none unsupported. Use low instead
GPT-6 Sol, GPT-6 Luna	Support none

Required changes

Area	Change
Endpoint	Use the Responses API for tool calling. Astra and 6.1 Sol support Chat Completions only for requests without tools. Sol and Luna support function calling in Chat Completions only with reasoning_effort: "none".
Sampling	When effort is not none, remove temperature, top_p, top_logprobs. Chat Completions: also remove logprobs. Responses: remove message.output_text.logprobs from include.
minimal effort	Replace with low. Compare on representative tasks.
none effort (Astra, 6.1 Sol)	Replace with low.
Prompt caching	Replace prompt_cache_retention with prompt_cache_options.ttl set to "30m" when migrating from GPT-5.5 or earlier. Review cache boundary and cache-write billing changes.
Model ID	Set model to the target ID. Astra and 6.1 Sol list one snapshot each: the bare ID.

OpenAI suggests Codex with the OpenAI Docs skill: $openai-docs migrate this project to the GPT-6 model family.

New API features

• Async tool calling: set async: true on a function or custom tool, then return the result later with the original call_id.
• Mid-turn steering: send user instructions while the model works.
• configuration_update input item: change reasoning effort mid- conversation without altering the prompt prefix, so caching survives.
• Fast and Ultrafast modes. Data residency differs: Astra Fast mode has no EU residency and no latency SLA.
• Carried over from GPT-5.6: computer use, Structured Outputs, streaming, Programmatic Tool Calling, multi-agent orchestration, prompt caching, persisted reasoning, compaction, pro mode.

Prompt and behavior tuning (optional)

• Initiative: Astra asks clarifying questions more than earlier models. Say to act when the request implies authorization, and to ask for approval once a concrete result is ready.
• Instruction priority: Astra weighs AGENTS.md and skill files heavily. Audit them. State that user instructions override skill guidance.
• Style: Astra favors long, formatted answers with lists and tables. State the style you want if you need prose.
• Subagents: Astra may delegate less than you want. Say when and how.
• Testing: Astra tests thoroughly, sometimes beyond what a small change needs. Calibrate the scope.
• Carry-over from GPT-5.5 guidance: outcome-first prompts, drop legacy scaffolding, keep few-shot examples for format calibration.

Upgrade paths

From	Notes
GPT-5.5	Apply the required changes above. Re-baseline effort: Astra defaults are not documented, 6.1 Sol defaults to medium. See gpt-5-5.md for the earlier surface.
GPT-5.x, GPT-4o, o-series	Apply the gpt-5-5.md steps first (Responses API, effort, structured outputs), then this guide.
GPT-6 Sol	Review the migration notes before moving to 6.1 Sol.

Search results also mention a GPT-5.6 model page. Treat GPT-5.6 as an intermediate release and verify its status on the OpenAI models page before targeting it.

Model ID reference

Version	Model ID	Status
GPT-6 Astra	gpt-6-astra	Current
GPT-6.1 Sol	gpt-6.1-sol	Current
GPT-6 Sol	gpt-6-sol	Existing
GPT-6 Luna	gpt-6-luna	Current
GPT-5.5	gpt-5.5	Previous generation (see gpt-5-5.md)

This site is open source. Improve this page.