BlogLLM Benchmark Scores as Published, Side by Side
GPT-5.4: Benchmark Scores as Published
Every result the catalog holds for GPT-5.4: 100 tests in 15 areas, from 12 pages. 67 printed by OpenAI itself, the rest by the company named beside each number. Each number is as printed; nothing is ranked or rescored.
Developer OpenAIReleased 2026-03-05Context window 1.05MMax output 128KInput Text, imageOutput TextWeights ClosedData version 2026103001, as of 2026-10-30
Knowledge Work
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| APEX-Agents | 33.3% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: xhigh |
| GDPval-AA (Elo) | 1674 | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: xhigh |
| OfficeQA Pro | 53.2% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Effort: xhigh |
| Vals Finance Agent v1.1 | 56% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh |
| Vending-Bench 2 (final balance in USD) | $6144.18 | Z.ai, GLM-5.1 model cardchecked 2026-10-15 | - |
| OfficeQA | 68.1% | OpenAI, Introducing GPT-5.4checked 2026-10-15 | Effort: xhigh |
| GDPval (wins or ties) | 83% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh |
| Investment Banking Modeling Tasks (Internal) | 87.3% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh |
Agentic Coding
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| SWE-Bench Pro | 57.7% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 8 more pages | Effort: xhigh |
| Terminal-Bench 2.0 | 75.1% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 7 more pages | Effort: xhigh |
| NL2Repo-Bench | 41.3 | Z.ai, GLM-5.1 model cardchecked 2026-10-15 | - |
| Multi-SWE-bench | 49% | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
| Expert-SWE (Internal) | 68.5% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Effort: xhigh |
ML Engineering
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| MLE-bench Lite (medal rate) | 71.2% | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
Science and Math
Long Context
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| GraphWalks BFS 1M (F1) | 9.4% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Effort: xhigh; Context: 1M |
| GraphWalks BFS up to 128k (F1) | 93% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 0K-128K |
| GraphWalks BFS 256k (F1) | 62.5% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Effort: xhigh; Context: 256k |
| MRCR v2 8-needle 128K-256K | 79.3% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 2 more pages | Effort: xhigh; Context: 128K-256K |
| MRCR v2 8-needle 512K-1M | 36.6% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 512K-1M |
| MRCR v2 8-needle 64K-128K | 86% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 2 more pages | Effort: xhigh; Context: 64K-128K |
| MRCR v2 8-needle 256K-512K | 57.5% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 256K-512K |
| GraphWalks BFS 256k to 1M (F1) | 21.4% | OpenAI, Introducing GPT-5.4checked 2026-10-15 | Effort: xhigh; Context: 256K-1M |
| GraphWalks parents up to 128k (accuracy) | 89.8% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 0-128K |
| GraphWalks parents 1M (F1) | 44.4% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Effort: xhigh; Context: 1M |
| MRCR v2 8-needle 16K-32K | 97.2% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 16K-32K |
| MRCR v2 8-needle 32K-64K | 90.5% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 32K-64K |
| MRCR v2 8-needle 4K-8K | 97.3% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 4K-8K |
| MRCR v2 8-needle 8K-16K | 91.4% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 8K-16K |
| GraphWalks parents 256k (F1) | 82.8% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Effort: xhigh; Context: 256k |
| GraphWalks parents 256k to 1M (accuracy) | 32.4% | OpenAI, Introducing GPT-5.4checked 2026-10-15 | Effort: xhigh; Context: 256K-1M |
Computer Use
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| OSWorld-Verified | 75% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 4 more pages | Effort: xhigh |
Multimodal Understanding
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| CharXiv Reasoning (no tools) | 82.8% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: xhigh; Tools: none |
| MMMU-Pro | 81.2% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 3 more pages | Effort: xhigh; Tools: none |
| MathVision | 92% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: xhigh; Tools: none |
| CharXiv Reasoning (with tools) | 90% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: xhigh; Tools: python |
| BabyVision (with tools) | 80.2% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: xhigh; Tools: python |
| MMMU-Pro (with tools) | 82.1% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 3 more pages | Effort: xhigh; Tools: with tools |
| BabyVision (no tools) | 49.7% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: xhigh; Tools: none |
| MathVision (with tools) | 96.1% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: xhigh; Tools: python |
| OmniDocBench 1.5 (overall edit distance)lower is better | 0.109 | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: none; Tools: none |
| V* Bench (with tools) | 98.4% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: xhigh; Tools: python |
Cybersecurity
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| CyberGym | 79% | OpenAI, Introducing GPT-5.5checked 2026-10-15; printed on 2 more pages | Effort: xhigh |
| Capture the Flag challenges | 83.7% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Effort: xhigh |
General Knowledge and Reasoning
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Humanity's Last Exam (no tools) | 39.8% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 5 more pages | Effort: xhigh; Tools: none |
| MMLU-Pro | 87.5% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: xhigh |
| Humanity's Last Exam (with tools) | 52.1% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 5 more pages | Effort: xhigh; Tools: with tools |
| SimpleQA Verified | 45.3% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: xhigh |
| ARC-AGI-2 | 73.3% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh |
| ARC-AGI-1 | 93.7% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh |
| Chinese SimpleQA | 76.8% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: xhigh |
Code Generation
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Codeforces rating | 3168 | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: xhigh |
Tool Use and Workflows
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Toolathlon | 54.6% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 6 more pages | Effort: xhigh |
| τ²-bench Telecom | 64.3% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 2 more pages | Effort: none |
| 98.9% | OpenAI, Introducing GPT-5.4checked 2026-10-15 | Effort: xhigh | |
| MCP Atlas | 67.2% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 3 more pages | Effort: xhigh |
| MCP Atlas (public set) | 67.2% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15; printed on 1 more page | Effort: xhigh |
| MCPMark | 62.5% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: xhigh |
| Claw-Eval pass^3 | 60.3% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: xhigh |
| τ³-bench (domain not stated) | 72.9% | Z.ai, GLM-5.1 model cardchecked 2026-10-15 | - |
| MM-ClawBench | 73.6% | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
| Claw-Eval pass@3 | 78.4% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: xhigh |
Web Search and Research
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| BrowseComp | 82.7% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 4 more pages | Effort: xhigh |
| DeepSearchQA (accuracy) | 63.7% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: xhigh; Tools: search, code interpreter, browsing |
| DeepSearchQA (F1) | 78.6% | Moonshot AI, Kimi K2.6 model cardchecked 2026-10-15 | Effort: xhigh; Tools: search, code interpreter, browsing |
Biosecurity
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| DNA sequence design for transcription factor binding (pass@1) | 12.82% | OpenAI, GPT-6 Astra System Cardchecked 2026-10-25 | Attempt budget: pass@1 |
| Hard-negative protein binding prediction (pass@4) | 3.5% | OpenAI, GPT-6 Astra System Cardchecked 2026-10-25 | Attempt budget: pass@4 |
Safety and Alignment
Preference and Composite Rankings
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Artificial Analysis Intelligence Index | 57 | MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15 | - |
Price
API list prices in US dollars per one million tokens, as printed on the developer's pricing page on the day checked.
| Price | Value | Printed by | Setup as printed |
|---|---|---|---|
| Input | $2.5 | OpenAI, OpenAI API pricingchecked 2026-10-15 | Context: up to 272,000 tokens; Long context: $5 per 1M tokens above 272,000 tokens |
| Cached input | $0.25 | OpenAI, OpenAI API pricingchecked 2026-10-15 | Context: up to 272,000 tokens; Long context: $0.5 per 1M tokens above 272,000 tokens |
| Output | $15 | OpenAI, OpenAI API pricingchecked 2026-10-15 | Context: up to 272,000 tokens; Long context: $22.5 per 1M tokens above 272,000 tokens |
| Input, batch | $1.25 | OpenAI, OpenAI API pricingchecked 2026-10-15 | Context: up to 272,000 tokens; Long context: $2.5 per 1M tokens above 272,000 tokens |
| Cached input, batch | $0.13 | OpenAI, OpenAI API pricingchecked 2026-10-15 | Context: up to 272,000 tokens; Long context: $0.25 per 1M tokens above 272,000 tokens |
| Output, batch | $7.5 | OpenAI, OpenAI API pricingchecked 2026-10-15 | Context: up to 272,000 tokens; Long context: $11.25 per 1M tokens above 272,000 tokens |
| Input, flex | $1.25 | OpenAI, OpenAI API pricingchecked 2026-10-15 | Context: up to 272,000 tokens; Long context: $2.5 per 1M tokens above 272,000 tokens |
| Cached input, flex | $0.13 | OpenAI, OpenAI API pricingchecked 2026-10-15 | Context: up to 272,000 tokens; Long context: $0.25 per 1M tokens above 272,000 tokens |
| Output, flex | $7.5 | OpenAI, OpenAI API pricingchecked 2026-10-15 | Context: up to 272,000 tokens; Long context: $11.25 per 1M tokens above 272,000 tokens |
| Input, fast | $5 | OpenAI, OpenAI API pricingchecked 2026-10-15 | - |
| Cached input, fast | $0.5 | OpenAI, OpenAI API pricingchecked 2026-10-15 | - |
| Output, fast | $30 | OpenAI, OpenAI API pricingchecked 2026-10-15 | - |
More from OpenAI
- GPT-6.1 Terra
- GPT-6.1 Sol
- GPT-6 Luna
- GPT-6 Sol
- GPT-6 Astra
- GPT-5.6 Luna
- GPT-5.6 Sol
- GPT-5.6 Terra
- GPT-5.5
- GPT-5.5 Pro
- GPT-5.4 mini
- GPT-5.4 nano
- GPT-5.4 Pro
- GPT-5.3-Codex
- GPT-5.2-Codex
- GPT-5.2
- GPT-5.2 Instant
- GPT-5.2 Pro
- GPT-5.1-Codex-Max
- GPT-5.1
- GPT-5.1 Instant
- GPT-5-Codex
- GPT-5
- GPT-5 mini
- GPT-5 Pro
- gpt-oss-120b
- gpt-oss-20b
- ChatGPT agent
- codex-1
- o3
- OpenAI o4-mini
- GPT-4.1
- GPT-4.5
- Deep research
- o3-mini
- o1
- o1-mini
- GPT-4o
- GPT-5.5 Instant
- gpt-5.6-cyber
Every number here is read from the same catalog file as the benchmark table, data version 2026103001. Where one page printed the model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.