GPT-6 Astra API

Price check: official, OpenRouter and Wokey

USD per million tokens. Official and OpenRouter rates were checked by hand on 2026-09-29, excluding tax; the Wokey rate is read from the live price list.

Checked 2026-09-29 · Official pricing page · OpenRouter endpoint list

GPT-6 Astra at a glance

Release date

2026-09-03 for selected API organizations and Daybreak; from 09-04 ChatGPT Business, Pro and Plus followed, with Enterprise enabled by admins. Pro, Business and Enterprise also get Astra Pro

Positioning

OpenAI pitches it “for our highest level of capability”. GPT-6 Sol (released 09-22) targets demanding tasks and GPT-6 Luna targets scale, and OpenAI says Astra remains its best model across the board

Context and output

1,050,000 context (922K max input) and 128K max output; text and image input, text output; knowledge cutoff 2026-04-30

Reasoning effort

low / medium / high / xhigh / max, with no none; configuration_update items change effort mid-conversation without breaking the cache prefix

Official price

$10 input / $50 output per million tokens, $1 cache read, $12.50 cache write; requests above 272K input are billed in full at 2x input and cache and 1.5x output ($20 / $2 / $25 / $75); Batch and Flex 50% off, Fast mode 2x

Cybersecurity limits

OpenAI rates its cyber capability as Critical under the Preparedness Framework. The public model refuses to write proof-of-concept exploits and is limited to secure code review and patching; full capability is offered through Daybreak

Codex

Codex adds context save and retrieval: notes carried across context windows, with earlier windows searchable. It is an experimental opt-in that will become the default; ChatGPT usage counts toward subscription allowances

Model ID and platforms

gpt-6-astra (the only snapshot); Chat Completions, Responses and Batch, with tool calling requiring Responses; also on AWS Bedrock and Microsoft Foundry; Tier 1 limits 500 RPM / 500K TPM

Checked 2026-09-29 · Sources: OpenAI · Migration guide · Benchmarks (Anthropic)

Official benchmarks: GPT-6 Astra vs Opus 5.5 vs Fable 5.1 vs Opus 5

From Anthropic’s Opus 5.5 launch page, keeping only the benchmarks that list a GPT-6 Astra score. These are vendor-reported results, and GPT-6 Astra scores are OpenAI-reported.

Anthropic compiled this table, and its benchmark selection leans toward Claude’s strengths. OpenAI’s launch reported a different set: 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3 and 100% on ExploitBench (GPT-5.6 Sol: 78.5%), which cannot be compared with the table directly.

Which to use: GPT-6 Astra, Claude Opus 5.5 or GPT-6 Sol

vs Claude Opus 5.5 / Fable 5.1

Opus 5.5 lists at $4 / $20, two-fifths of Astra’s price, and clearly leads on Terminal-Bench 4.0, GDPval-AA and Humanity’s Last Exam above; Astra leads on AutomationBench and Terminal-Bench-Science. On Artificial Analysis’ Intelligence Index Opus 5.5 scores 58 at max for $5.98 per task and Astra 53 for $3.26: Astra costs more per token but uses far fewer output tokens, so each task costs less. Fable 5.1 shares Astra’s $10 / $50 price and its score of 53, but costs $7.63 per task, more than twice as much.

vs GPT-6 Sol / Luna

GPT-6 Sol lists at $2 / $10, a fifth of Astra’s price, and GPT-6 Luna at $0.10 / $0.50. On Artificial Analysis’ Intelligence Index Sol scores 48 for $1.06 per task and Luna 37 for $0.07; on its Coding Agent Index Astra scores 62, Sol 57 and Luna 41. Start everyday coding and routine agent work on Sol and move to Astra for problems Sol cannot solve; use Luna for high-volume classification and extraction.

Choosing effort

Artificial Analysis measured low at 46 for $0.82 per task, medium at 50 for $1.54, high at 51 for $1.73 and max at 53 for $3.26, with every level on the cost-versus-score frontier. Start at medium: high costs about 12% more for 1 point, and max costs more than twice as much for 3 points. When migrating, replace none with low and start evaluating minimal at low.

OpenAI migration guide

Third-party and community reviews

• Artificial Analysis: It scores 53 on the Intelligence Index at max, tied for first with Fable 5.1 and 6 points above GPT-5.6 Sol; a task costs $3.26, about 40% of Fable 5.1’s $7.63 and about 60% more than GPT-5.6 Sol. It uses about 27K output tokens per task, a third of Fable 5.1’s 78K. It scores 62 on the Coding Agent Index, level with Fable 5.1 and ahead of Opus 5 (60) and GPT-5.6 Sol (55), at $7.09 per task, about 40% cheaper than Fable 5.1 and 30% cheaper than Opus 5. Its AA-Omniscience hallucination rate fell to 51% from GPT-5.6 Sol’s 92%. It trails GPT-5.6 Sol by about 45 Elo on GDPval-AA but needs only 24 turns per task, versus 45 for GPT-5.6 Sol and 60 for Fable 5.1 and Opus 5. Every Wokey meter is 9% of the official rate, so the same task comes to about $0.29 at max and $0.14 at medium.
• OpenAI system card and safety testing: In agentic tests without a confirmation policy, 3.4% of runs ended in misaligned outcomes versus 18.8% for GPT-5.6 Sol, and Gray Swan’s indirect prompt injection attacks succeeded 8.5% of the time versus 27.0%. A 100% ExploitBench score led OpenAI to rate a model’s cyber capability as Critical for the first time; the public model is limited to secure code review and patching, and Daybreak access will expand later. These figures are OpenAI-reported.
• Hacker News (2,279-point thread): An external Codex tester called it a “grounded collaborator and executor” that avoids doing work you did not ask for. Others found Fable better at inferring intent and said Astra needs very specific instructions, while some liked that it does not treat questions as instructions. After general availability, commenters reported usage at about 2.5x the rate of GPT-5.6 Sol, and one used up a 5-hour limit in 15 messages.

Migrating from GPT-5.6 Sol and older models: parameters to remove and behavior changes

These changes come from OpenAI’s model page and latest-model guide. Check your request parameters, cache settings, tool calling and prompts before switching the model to gpt-6-astra.

• Remove temperature, top_p and top_logprobs; on Chat Completions also remove logprobs, and on Responses drop message.output_text.logprobs from include. Astra owns its sampling policy and rejects requests with these fields. Through Wokey the gateway strips temperature, top_p, top_logprobs and logprobs before forwarding.
• reasoning.effort does not accept none: replace none with low, and start evaluating minimal at low. Through Wokey none and minimal are mapped to low automatically, so the request does not fail.
• Chat Completions can call Astra, but tool calling requires Responses, so move tool-using agents to /v1/responses. Through Wokey, Chat Completions requests are converted to Responses before forwarding, so Chat requests with tools also work.
• Caching: replace prompt_cache_retention with prompt_cache_options.ttl: "30m", and to change effort mid-session send a configuration_update item instead of editing earlier input, so the cache prefix stays valid.
• Above 272K input tokens the whole request bills input and cache at 2x and output at 1.5x, so long-context costs jump suddenly; compaction or retrieval can keep input under 272K. Through Wokey there is no tier, and the whole 1M context bills at one rate.
• Behavior changes: it asks clarifying questions and pauses for confirmation, delegates less and tests more broadly; formatting is more detailed and phrases repeat. It is more sensitive to skills and context files, so OpenAI recommends auditing existing skills and turning them into minimal routers, since over-specific guidance can hold it back.
• Cybersecurity requests: the public model refuses to write proof-of-concept exploits, so penetration testing and exploit reproduction tasks may be refused outright, while secure code review and patching work normally. Full capability requires Daybreak access.
• In Codex, run $openai-docs migrate this project to GPT-6 Astra to update the project against the official checklist. Fast mode has no latency SLA, and EU data residency supports Standard processing only.

Request body after migration (/v1/responses)

{
  "model": "gpt-6-astra",
  "reasoning": { "effort": "medium" },
  "input": [{ "role": "user", "content": "Find and fix the flaky test in this repo" }]
}

Catalog pricing snapshot

Prices are per 1M tokens; request-time pricing controls settlement.

Official price source: https://developers.openai.com/api/docs/models/gpt-6-astra

Model capabilities

Context

1,050,000 tokens

Max output

128,000 tokens

Streaming

Supported

Tools

Supported

Vision

Supported

Supported API forms

/v1/chat/completions · /v1/responses

Send a request

/v1/chat/completions

curl https://api.wokey.ai/v1/chat/completions \
  -H "Authorization: Bearer $WOKEY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-6-astra","messages":[{"role":"user","content":"Hello"}]}'

/v1/responses

curl https://api.wokey.ai/v1/responses \
  -H "Authorization: Bearer $WOKEY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-6-astra","input":"Hello"}'

Use GPT-6 Astra in your tools

The gateway is https://api.wokey.ai (OpenAI-compatible clients usually take https://api.wokey.ai/v1), and the model ID is gpt-6-astra. Open the guide for the client you use.

• Use GPT-6 Astra in Codex CLI: OpenAI Responses · base URL and API key setup
• Use GPT-6 Astra in opencode: OpenAI Chat Completions · base URL and API key setup
• Use GPT-6 Astra in OpenClaw: OpenAI Chat Completions · base URL and API key setup
• Use GPT-6 Astra in Hermes Agent: OpenAI Chat Completions · base URL and API key setup
• OpenRouter alternative: Wokey vs OpenRouter: Compare token prices, payment fees, API forms, and upstream sources.
• Verifiable AI API: check a GPT-6 Astra response came from the official upstream: Which responses carry a tee.proof, and what a proof does and does not show.
• Verify a GPT-6 Astra response online: AI API relay verifier: Paste a response with its tee.proof and check it locally in your browser; nothing is uploaded.

Example production call

These usage values come from a completed GPT-6 Astra call that succeeded on its first attempt and was reconciled with its bill.

Total input

675347

Output

669

Cached share of input

99.1%

Call duration

23.74 s

Responses usage

Compiled from the billing record into API field examples. input_tokens is total input, including cached tokens.

{
  "model": "gpt-6-astra",
  "usage": {
    "input_tokens": 675347,
    "input_tokens_details": {
      "cached_tokens": 669312
    },
    "output_tokens": 669,
    "total_tokens": 676016
  }
}

Token accounting

These are the final per-million-token rates recorded for this call.

Billed amount

$0.068680

Cost = sum of tokens × their rate ÷ 1,000,000, rounded to six decimals after summing.

Source of this call

This call used the OpenAI channel at chatgpt.com with OAuth authorization. Usage was metered upstream and the request returned HTTP 200.

Verify Prompt Cache

1. Keep the shared prefix of long prompts identical and reuse context according to the selected API’s cache rules.
2. Check input_tokens_details.cached_tokens; zero means no cache reads were recorded for this call.
3. Uncached input = input_tokens minus cached tokens. Price each bucket separately and add the output charge.

Frequently asked questions

Related models and documentation

• GPT-6.1 Sol
• GPT-6 Sol
• GPT-6 Luna
• GPT-5.6 Sol
• API documentation
• All models and pricing
• Provider Node