GPT-6 Astra
openai/gpt-6-astra-20260903
OpenAI's GPT-6 Astra is a multimodal model that accepts file, image, and text inputs and generates text outputs. It features a 1,050,000-token context window, supports tool use and reasoning, and is a proprietary model released in September 2026.
- tool use
- structured outputs
- reasoning
At a glance
Specifications
What the API gives you
- Context window
- 1.1M tokens
- Max output
- 128K tokens
- Input price
- $10 / 1M tokens
- Output price
- $50 / 1M tokens
- Cache read
- $1 / 1M tokens
- Modalities
- file, image, text → text
- Knowledge cutoff
- —
- Released
- Sep 3, 2026
Measured evidence
Public benchmark record
- Frontier-Bench‡
- 58.2%2026-09-03Codex, max effort
- LiveBench Agentic Coding
- 57.3%2026-06-25
- LiveBench Coding
- 80.4%2026-06-25
- LiveBench Data Analysis
- 83.0%2026-06-25
- LiveBench Instruction Following
- 75.6%2026-06-25
- LiveBench Language
- 89.4%2026-06-25
- LiveBench Math
- 96.8%2026-06-25
- LiveBench Reasoning
- 92.7%2026-06-25
- LMArena Creative Writing
- 1431 Elo2026-09-11
- LMArena Creative Writing
- 1429 Elo2026-09-13
- LMArena Instruction Following
- 1438 Elo2026-09-11
- LMArena Instruction Following
- 1439 Elo2026-09-13
- LMArena Long Query
- 1450 Elo2026-09-11
- LMArena Long Query
- 1453 Elo2026-09-13
- LMArena Text
- 1478 Elo2026-09-11
- LMArena Text
- 1480 Elo2026-09-13
- LMArena Vision
- 1284 Elo2026-09-13
- LMArena WebDev
- 1797 Elo2026-09-05
- LMArena WebDev
- 1800 Elo2026-09-11
- LMArena WebDev
- 1793 Elo2026-09-22
- LMArena WebDev
- 1792 Elo2026-09-23
- Terminal-Bench 2.1‡
- 58.2%2026-09-03max effort
LMArena data licensed CC-BY-4.0. Aider (Apache-2.0) and SWE-bench (MIT) are open. Design Arena uses blind human preference. ‡ Submitted scaffold plus model result. Compare the harness and effort settings, not only the model name.
Task evidence
How broad is the support?
Each status is exact to this model version and the selected task. Official release claims remain provisional.
- Best LLM for Business WritingEstablished evidence
100% of intended weighted task evidence across 2 independent sources. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for Marketing CopyEstablished evidence
100% of intended weighted task evidence across 2 independent sources. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for Chat & RoleplayEstablished evidence
100% of intended weighted task evidence across 2 independent sources. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for Creative WritingEstablished evidence
100% of intended weighted task evidence across 2 independent sources. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for WritingEstablished evidence
100% of intended weighted task evidence across 2 independent sources. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for Roleplay & CharacterEstablished evidence
100% of intended weighted task evidence across 2 independent sources. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for a General AssistantEstablished evidence
100% of intended weighted task evidence across 2 independent sources. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for Logic & Problem SolvingEstablished evidence
85% of intended weighted task evidence across 2 independent sources. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for Code CompletionEstablished evidence
81% of intended weighted task evidence across 2 independent sources. 4 of 4 configured evaluation feeds currently have results.
- Best LLM for Competition MathEstablished evidence
75% of intended weighted task evidence across 2 independent sources. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for Math & ReasoningEstablished evidence
75% of intended weighted task evidence across 2 independent sources. 5 of 5 configured evaluation feeds currently have results.
- Best Open-Weight LLM for Self-HostingEstablished evidence
67% of intended weighted task evidence across 2 independent sources. 3 of 4 configured evaluation feeds currently have results.
- Best LLM for Image UnderstandingEstablished evidence
65% of intended weighted task evidence across 2 independent sources. 3 of 4 configured evaluation feeds currently have results.
- Best LLM for Fast SummariesEstablished evidence
64% of intended weighted task evidence across 2 independent sources. 3 of 4 configured evaluation feeds currently have results.
- Best Value LLMEstablished evidence
64% of intended weighted task evidence across 2 independent sources. 3 of 4 configured evaluation feeds currently have results.
- Best LLM for Autonomous AgentsEstablished evidence
63% of intended weighted task evidence across 3 independent sources. 7 of 7 configured evaluation feeds currently have results.
- Best LLM for Document SummarizationEstablished evidence
60% of intended weighted task evidence across 2 independent sources. 4 of 6 configured evaluation feeds currently have results.
- Best LLM for SummarizationEstablished evidence
60% of intended weighted task evidence across 2 independent sources. 4 of 6 configured evaluation feeds currently have results.
- Best LLM for Literary TranslationProvisional evidence
55% of intended weighted task evidence across 2 independent sources; broader independent confirmation is still missing from WMT translation. 4 of 5 configured evaluation feeds currently have results.
- Best LLM for Everyday TranslationProvisional evidence
55% of intended weighted task evidence across 2 independent sources; broader independent confirmation is still missing from WMT translation. 4 of 5 configured evaluation feeds currently have results.
- Best LLM for TranslationProvisional evidence
55% of intended weighted task evidence across 2 independent sources; broader independent confirmation is still missing from WMT translation. 4 of 5 configured evaluation feeds currently have results.
- Best LLM for Agentic CodingProvisional evidence
55% of intended weighted task evidence across 3 independent sources; broader independent confirmation is still missing from Design Arena Full-stack or Design Arena Web-apps. 8 of 8 configured evaluation feeds currently have results. Official release evidence is also tracked provisionally.
- Best LLM for CodingProvisional evidence
54% of intended weighted task evidence across 3 independent sources; broader independent confirmation is still missing from Aider Polyglot or Design Arena Coding. 8 of 8 configured evaluation feeds currently have results. Official release evidence is also tracked provisionally.
- Best Cheap LLM for High VolumeProvisional evidence
50% of intended weighted task evidence across 2 independent sources; broader independent confirmation is still missing from Route reliability. 3 of 4 configured evaluation feeds currently have results.
- Cheapest Capable LLMProvisional evidence
50% of intended weighted task evidence across 2 independent sources; broader independent confirmation is still missing from Route reliability. 3 of 4 configured evaluation feeds currently have results.
- Best LLM for Massive ContextProvisional evidence
50% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LongBench v2. 3 of 4 configured evaluation feeds currently have results.
- Best Local LLM for CodingProvisional evidence
50% of intended weighted task evidence across 2 independent sources; broader independent confirmation is still missing from Aider Polyglot or SWE-rebench. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for OCR & DocumentsProvisional evidence
45% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Structured-output eval or VLMEvalKit tasks. 3 of 5 configured evaluation feeds currently have results.
- Best LLM for Vision & DocumentsProvisional evidence
45% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Structured-output eval or VLMEvalKit tasks. 3 of 5 configured evaluation feeds currently have results.
- Best LLM for Charts & DiagramsProvisional evidence
45% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Structured-output eval or VLMEvalKit tasks. 3 of 5 configured evaluation feeds currently have results.
- Best LLM for Document ParsingProvisional evidence
45% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Structured-output eval or VLMEvalKit tasks. 3 of 5 configured evaluation feeds currently have results.
- Best LLM for Debugging & Code ReviewProvisional evidence
40% of intended weighted task evidence across 2 independent sources; broader independent confirmation is still missing from Aider Polyglot or SWE-bench Verified. 6 of 6 configured evaluation feeds currently have results. Official release evidence is also tracked provisionally.
- Best LLM for Frontend & UIProvisional evidence
40% of intended weighted task evidence across 2 independent sources; broader independent confirmation is still missing from Design Arena Coding or Design Arena UI. 5 of 5 configured evaluation feeds currently have results. Official release evidence is also tracked provisionally.
- Best LLM for JSON & Schema OutputProvisional evidence
39% of intended weighted task evidence across 2 independent sources; broader independent confirmation is still missing from Berkeley Function Calling or Structured-output eval. 4 of 5 configured evaluation feeds currently have results.
- Best LLM for Structured Data ExtractionProvisional evidence
39% of intended weighted task evidence across 2 independent sources; broader independent confirmation is still missing from Berkeley Function Calling or Structured-output eval. 4 of 5 configured evaluation feeds currently have results.
- Best LLM for RAG & Long ContextProvisional evidence
35% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from FACTS Grounding or LMArena Search. 4 of 6 configured evaluation feeds currently have results.
- Best LLM for AI AgentsProvisional evidence
33% of intended weighted task evidence across 2 independent sources; broader independent confirmation is still missing from Berkeley Function Calling or LMArena Agent. 7 of 7 configured evaluation feeds currently have results. Official release evidence is also tracked provisionally.
- Best LLM for High-Volume ExtractionProvisional evidence
31% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Route reliability or Structured-output eval. 2 of 4 configured evaluation feeds currently have results.
- Best LLM for Hermes AgentProvisional evidence
24% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Berkeley Function Calling or LMArena Agent. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for OpenHandsProvisional evidence
24% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Berkeley Function Calling or LMArena Agent. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for GooseProvisional evidence
24% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Berkeley Function Calling or LMArena Agent. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for OpenClawProvisional evidence
24% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Berkeley Function Calling or LMArena Agent. 5 of 5 configured evaluation feeds currently have results.
- Best LLM for Tool & Function CallingProvisional evidence
10% of intended weighted task evidence across 0 independent sources; broader independent confirmation is still missing from Berkeley Function Calling or LMArena Agent. 4 of 5 configured evaluation feeds currently have results.
- Best LLM for Computer & Browser UseProvisional evidence
10% of intended weighted task evidence across 0 independent sources; broader independent confirmation is still missing from Berkeley Function Calling or LMArena Agent. 4 of 4 configured evaluation feeds currently have results. Official release evidence is also tracked provisionally.
- Best LLM for Grounded RAGProvisional evidence
10% of intended weighted task evidence across 0 independent sources; broader independent confirmation is still missing from CRAG or FACTS Grounding. 2 of 5 configured evaluation feeds currently have results.
Official release evidence · provisional
What the launch material actually says
Verbatim claims from the model provider. These help on launch day but do not count as independent confirmation.
“GPT‑6 Astra is our best model for adhering to existing templates and producing slides that are well laid out”
OpenAI · GPT-6 Astra: A new generation of intelligence | OpenAI“GPT‑6 Astra reaches a new high in the comparison shown, scoring 59.3%, compared with 55.5% for Claude Opus 5 and 53.6% for GPT‑5.6 Sol.”
OpenAI · GPT-6 Astra: A new generation of intelligence | OpenAI“Astra also saturates ARC-AGI-3 with a 99.9% score”
OpenAI · GPT-6 Astra: A new generation of intelligence | OpenAI“GPT‑6 Astra delivers state-of-the-art performance on our internal coding benchmarks”
OpenAI · GPT-6 Astra: A new generation of intelligence | OpenAI
Recommendation coverage
Where this model appears
Catalog data via OpenRouter and models.dev. Last checked within the past 6 hours.