Leaderboards
One board per question
No single ranking answers every question. Pick the measurement that matches the work you are actually doing — a model that tops GPQA may be nowhere near the top on long-context retrieval.
Headline
The composite indices
Aggregate scores. Useful for a shortlist, misleading as a final answer.
Intelligence Index
index · higher is better
- Claude Opus 5Anthropic63.1
- Claude Fable 5Anthropic62.1
- GPT-5.6 SolOpenAI60.9
- Kimi K3Moonshot AI59.7
- Qwen3.8 MaxQwen58.1
- Claude Opus 4.8Anthropic57.3
- Muse Spark 1.2Meta56.8
- GPT-5.6 TerraOpenAI56.6
- GPT-5.5OpenAI56.3
- Grok 4.5xAI55.8
Composite of ten reasoning, knowledge, coding and agentic evaluations.
View as table
| Model | index |
|---|---|
| Claude Opus 5 (Anthropic) | 63.1 |
| Claude Fable 5 (Anthropic) | 62.1 |
| GPT-5.6 Sol (OpenAI) | 60.9 |
| Kimi K3 (Moonshot AI) | 59.7 |
| Qwen3.8 Max (Qwen) | 58.1 |
| Claude Opus 4.8 (Anthropic) | 57.3 |
| Muse Spark 1.2 (Meta) | 56.8 |
| GPT-5.6 Terra (OpenAI) | 56.6 |
| GPT-5.5 (OpenAI) | 56.3 |
| Grok 4.5 (xAI) | 55.8 |
Coding Index
index · higher is better
- Claude Opus 5Anthropic78.0
- GPT-5.6 SolOpenAI77.4
- GPT-5.6 TerraOpenAI76.7
- Claude Fable 5Anthropic76.5
- Kimi K3Moonshot AI76.2
- GPT-5.5OpenAI74.9
- Claude Opus 4.8Anthropic74.3
- Claude Opus 4.7Anthropic73.6
- Grok 4.5xAI72.4
- Muse Spark 1.2Meta72.2
SciCode, LiveCodeBench and Terminal-Bench Hard, combined.
View as table
| Model | index |
|---|---|
| Claude Opus 5 (Anthropic) | 78.0 |
| GPT-5.6 Sol (OpenAI) | 77.4 |
| GPT-5.6 Terra (OpenAI) | 76.7 |
| Claude Fable 5 (Anthropic) | 76.5 |
| Kimi K3 (Moonshot AI) | 76.2 |
| GPT-5.5 (OpenAI) | 74.9 |
| Claude Opus 4.8 (Anthropic) | 74.3 |
| Claude Opus 4.7 (Anthropic) | 73.6 |
| Grok 4.5 (xAI) | 72.4 |
| Muse Spark 1.2 (Meta) | 72.2 |
Agentic Index
index · higher is better
- Claude Opus 5Anthropic59.2
- Qwen3.8 MaxQwen58.4
- GPT-5.6 SolOpenAI57.8
- Claude Fable 5Anthropic56.6
- Kimi K3Moonshot AI54.3
- GPT-5.6 TerraOpenAI50.2
- Claude Sonnet 5Anthropic49.7
- Claude Opus 4.8Anthropic49.4
- Grok 4.5xAI48.9
- V4 Flash 0731DeepSeek48.4
Tool use, long-horizon planning and terminal work.
View as table
| Model | index |
|---|---|
| Claude Opus 5 (Anthropic) | 59.2 |
| Qwen3.8 Max (Qwen) | 58.4 |
| GPT-5.6 Sol (OpenAI) | 57.8 |
| Claude Fable 5 (Anthropic) | 56.6 |
| Kimi K3 (Moonshot AI) | 54.3 |
| GPT-5.6 Terra (OpenAI) | 50.2 |
| Claude Sonnet 5 (Anthropic) | 49.7 |
| Claude Opus 4.8 (Anthropic) | 49.4 |
| Grok 4.5 (xAI) | 48.9 |
| V4 Flash 0731 (DeepSeek) | 48.4 |
Output speed
tokens/s · higher is better
- V3 FastMorph3,502 t/s
- V3 LargeMorph3,129 t/s
- Apply 3Relace3,080 t/s
- gpt-oss-120bOpenAI773 t/s
- GLM 4.7Z.ai502 t/s
- 443 t/s
- Qwen3 32BQwen423 t/s
- gpt-oss-safeguard-20bOpenAI422 t/s
- 337 t/s
- M2.7MiniMax307 t/s
Median output tokens per second across serving providers.
View as table
| Model | tokens/s |
|---|---|
| V3 Fast (Morph) | 3,502 t/s |
| V3 Large (Morph) | 3,129 t/s |
| Apply 3 (Relace) | 3,080 t/s |
| gpt-oss-120b (OpenAI) | 773 t/s |
| GLM 4.7 (Z.ai) | 502 t/s |
| Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image) (Google) | 443 t/s |
| Qwen3 32B (Qwen) | 423 t/s |
| gpt-oss-safeguard-20b (OpenAI) | 422 t/s |
| Grok 4.20 Multi-Agent (xAI) | 337 t/s |
| M2.7 (MiniMax) | 307 t/s |
Context window
tokens · higher is better
- 2M
- Grok 4.20xAI2M
- Llama 4 ScoutMeta1.31M
- GPT-5.6 Luna ProOpenAI1.05M
- GPT-5.6 LunaOpenAI1.05M
- GPT-5.6 Terra ProOpenAI1.05M
- GPT-5.6 TerraOpenAI1.05M
- GPT-5.6 Sol ProOpenAI1.05M
- GPT-5.6 SolOpenAI1.05M
- GPT-5.5 ProOpenAI1.05M
Maximum tokens the model accepts in one request.
View as table
| Model | tokens |
|---|---|
| Grok 4.20 Multi-Agent (xAI) | 2M |
| Grok 4.20 (xAI) | 2M |
| Llama 4 Scout (Meta) | 1.31M |
| GPT-5.6 Luna Pro (OpenAI) | 1.05M |
| GPT-5.6 Luna (OpenAI) | 1.05M |
| GPT-5.6 Terra Pro (OpenAI) | 1.05M |
| GPT-5.6 Terra (OpenAI) | 1.05M |
| GPT-5.6 Sol Pro (OpenAI) | 1.05M |
| GPT-5.6 Sol (OpenAI) | 1.05M |
| GPT-5.5 Pro (OpenAI) | 1.05M |
Arena Elo
Dot position on a 1,325–1,475 scale
- Kimi K3Moonshot AI1,455
- Claude Fable 5Anthropic1,397
- Claude Opus 5Anthropic1,393
- GPT-5.6 SolOpenAI1,380
- GLM 5.2Z.ai1,363
- Gemini 3.6 FlashGoogle1,346
- Claude Opus 4.7Anthropic1,337
- Claude Sonnet 5Anthropic1,336
- GPT-5.5OpenAI1,336
- Gemini 3.1 Pro PreviewGoogle1,333
1,3251,475
Head-to-head generation quality, judged pairwise. Elo has no meaningful zero, so bars from a zero baseline would flatten every gap.
View as table
| Model | Elo |
|---|---|
| Kimi K3 (Moonshot AI) | 1,455 |
| Claude Fable 5 (Anthropic) | 1,397 |
| Claude Opus 5 (Anthropic) | 1,393 |
| GPT-5.6 Sol (OpenAI) | 1,380 |
| GLM 5.2 (Z.ai) | 1,363 |
| Gemini 3.6 Flash (Google) | 1,346 |
| Claude Opus 4.7 (Anthropic) | 1,337 |
| Claude Sonnet 5 (Anthropic) | 1,336 |
| GPT-5.5 (OpenAI) | 1,336 |
| Gemini 3.1 Pro Preview (Google) | 1,333 |
Individual evaluations
The measurements underneath
Each of these is a real test with its own failure modes. A model absent from a board was not run on it — that is not the same as scoring zero.
GPQA Diamond
Percentage correct · higher is better
- GPT-5.6 SolOpenAI94.1%
- Gemini 3.1 Pro PreviewGoogle94.1%
- Kimi K3Moonshot AI93.5%
- GPT-5.5OpenAI93.5%
- Claude Opus 5Anthropic93.2%
- Grok 4.5xAI93.1%
- MiniMax M3MiniMax92.9%
- Gemini 3.6 FlashGoogle92.8%
- Qwen3.8 MaxQwen92.7%
- Claude Fable 5Anthropic92.6%
Graduate-level science questions written to resist search.
View as table
| Model | Score |
|---|---|
| GPT-5.6 Sol (OpenAI) | 94.1% |
| Gemini 3.1 Pro Preview (Google) | 94.1% |
| Kimi K3 (Moonshot AI) | 93.5% |
| GPT-5.5 (OpenAI) | 93.5% |
| Claude Opus 5 (Anthropic) | 93.2% |
| Grok 4.5 (xAI) | 93.1% |
| MiniMax M3 (MiniMax) | 92.9% |
| Gemini 3.6 Flash (Google) | 92.8% |
| Qwen3.8 Max (Qwen) | 92.7% |
| Claude Fable 5 (Anthropic) | 92.6% |
Humanity's Last Exam
Percentage correct · higher is better
- Claude Fable 5Anthropic55.5%
- Claude Opus 5Anthropic54.9%
- GPT-5.6 SolOpenAI49.5%
- Claude Opus 4.8Anthropic48.7%
- Gemini 3.1 Pro PreviewGoogle47.0%
- Kimi K3Moonshot AI46.9%
- Muse Spark 1.1Meta46.2%
- GPT-5.5OpenAI45.8%
- Muse Spark 1.2Meta45.5%
- GPT-5.4OpenAI43.7%
A deliberately brutal cross-disciplinary exam; scores stay low.
View as table
| Model | Score |
|---|---|
| Claude Fable 5 (Anthropic) | 55.5% |
| Claude Opus 5 (Anthropic) | 54.9% |
| GPT-5.6 Sol (OpenAI) | 49.5% |
| Claude Opus 4.8 (Anthropic) | 48.7% |
| Gemini 3.1 Pro Preview (Google) | 47.0% |
| Kimi K3 (Moonshot AI) | 46.9% |
| Muse Spark 1.1 (Meta) | 46.2% |
| GPT-5.5 (OpenAI) | 45.8% |
| Muse Spark 1.2 (Meta) | 45.5% |
| GPT-5.4 (OpenAI) | 43.7% |
SciCode
Percentage correct · higher is better
- Claude Fable 5Anthropic60.2%
- Gemini 3.1 Pro PreviewGoogle58.9%
- Kimi K3Moonshot AI58.7%
- Muse Spark 1.1Meta58.2%
- GPT-5.4OpenAI56.6%
- Muse Spark 1.2Meta56.4%
- GPT-5.6 SolOpenAI56.1%
- GPT-5.5OpenAI56.1%
- Claude Opus 5Anthropic55.7%
- GPT-5.2-CodexOpenAI54.6%
Research-grade scientific coding problems.
View as table
| Model | Score |
|---|---|
| Claude Fable 5 (Anthropic) | 60.2% |
| Gemini 3.1 Pro Preview (Google) | 58.9% |
| Kimi K3 (Moonshot AI) | 58.7% |
| Muse Spark 1.1 (Meta) | 58.2% |
| GPT-5.4 (OpenAI) | 56.6% |
| Muse Spark 1.2 (Meta) | 56.4% |
| GPT-5.6 Sol (OpenAI) | 56.1% |
| GPT-5.5 (OpenAI) | 56.1% |
| Claude Opus 5 (Anthropic) | 55.7% |
| GPT-5.2-Codex (OpenAI) | 54.6% |
τ²-bench (Telecom)
Percentage correct · higher is better
- GLM 5.2Z.ai99.1%
- GLM 4.7 FlashZ.ai98.8%
- Claude Fable 5Anthropic98.5%
- Step 3.7 FlashStepFun98.5%
- GLM 5V TurboZ.ai98.5%
- GLM 5 TurboZ.ai98.5%
- GLM 5Z.ai98.2%
- Grok 4.3xAI97.7%
- GLM 5.1Z.ai97.7%
- Qwen3.6 PlusQwen97.7%
Tool-use and policy adherence in a simulated telecom support agent.
View as table
| Model | Score |
|---|---|
| GLM 5.2 (Z.ai) | 99.1% |
| GLM 4.7 Flash (Z.ai) | 98.8% |
| Claude Fable 5 (Anthropic) | 98.5% |
| Step 3.7 Flash (StepFun) | 98.5% |
| GLM 5V Turbo (Z.ai) | 98.5% |
| GLM 5 Turbo (Z.ai) | 98.5% |
| GLM 5 (Z.ai) | 98.2% |
| Grok 4.3 (xAI) | 97.7% |
| GLM 5.1 (Z.ai) | 97.7% |
| Qwen3.6 Plus (Qwen) | 97.7% |
Terminal-Bench Hard
Percentage correct · higher is better
- GPT-5.6 SolOpenAI65.9%
- Claude Fable 5Anthropic62.9%
- GPT-5.5OpenAI60.6%
- Claude Opus 4.8Anthropic58.3%
- GPT-5.6 TerraOpenAI57.6%
- GPT-5.4OpenAI57.6%
- Gemini 3.1 Pro PreviewGoogle53.8%
- GPT-5.3-CodexOpenAI53.0%
- GPT-5.4 MiniOpenAI52.3%
- Claude Opus 4.7Anthropic51.5%
Hard end-to-end tasks solved inside a real terminal.
View as table
| Model | Score |
|---|---|
| GPT-5.6 Sol (OpenAI) | 65.9% |
| Claude Fable 5 (Anthropic) | 62.9% |
| GPT-5.5 (OpenAI) | 60.6% |
| Claude Opus 4.8 (Anthropic) | 58.3% |
| GPT-5.6 Terra (OpenAI) | 57.6% |
| GPT-5.4 (OpenAI) | 57.6% |
| Gemini 3.1 Pro Preview (Google) | 53.8% |
| GPT-5.3-Codex (OpenAI) | 53.0% |
| GPT-5.4 Mini (OpenAI) | 52.3% |
| Claude Opus 4.7 (Anthropic) | 51.5% |
LiveCodeBench
Percentage correct · higher is better
- GLM 4.7Z.ai89.4%
- GPT-5.2OpenAI88.9%
- gpt-oss-120bOpenAI87.8%
- GPT-5.1OpenAI86.8%
- o4 Mini HighOpenAI85.9%
- o4 MiniOpenAI85.9%
- Kimi K2 ThinkingMoonshot AI85.3%
- GPT-5.1-CodexOpenAI84.9%
- GPT-5OpenAI84.6%
- GPT-5 MiniOpenAI83.8%
Contest programming on problems released after training.
View as table
| Model | Score |
|---|---|
| GLM 4.7 (Z.ai) | 89.4% |
| GPT-5.2 (OpenAI) | 88.9% |
| gpt-oss-120b (OpenAI) | 87.8% |
| GPT-5.1 (OpenAI) | 86.8% |
| o4 Mini High (OpenAI) | 85.9% |
| o4 Mini (OpenAI) | 85.9% |
| Kimi K2 Thinking (Moonshot AI) | 85.3% |
| GPT-5.1-Codex (OpenAI) | 84.9% |
| GPT-5 (OpenAI) | 84.6% |
| GPT-5 Mini (OpenAI) | 83.8% |
AA-LCR (long context)
Percentage correct · higher is better
- Muse Spark 1.2Meta83.3%
- Kimi K3Moonshot AI82.7%
- Muse Spark 1.1Meta81.3%
- Gemini 3.5 FlashGoogle81.0%
- MiniMax M3MiniMax80.3%
- GPT-5.6 TerraOpenAI79.7%
- GPT-5.2-CodexOpenAI79.3%
- GPT-5.2OpenAI79.3%
- Gemini 3.6 FlashGoogle79.0%
- GPT-5.5OpenAI79.0%
Long-context reasoning across large documents.
View as table
| Model | Score |
|---|---|
| Muse Spark 1.2 (Meta) | 83.3% |
| Kimi K3 (Moonshot AI) | 82.7% |
| Muse Spark 1.1 (Meta) | 81.3% |
| Gemini 3.5 Flash (Google) | 81.0% |
| MiniMax M3 (MiniMax) | 80.3% |
| GPT-5.6 Terra (OpenAI) | 79.7% |
| GPT-5.2-Codex (OpenAI) | 79.3% |
| GPT-5.2 (OpenAI) | 79.3% |
| Gemini 3.6 Flash (Google) | 79.0% |
| GPT-5.5 (OpenAI) | 79.0% |
AIME 2025
Percentage correct · higher is better
- GPT-5.2OpenAI99.0%
- GPT-5.1-CodexOpenAI95.7%
- GLM 4.7Z.ai95.0%
- Kimi K2 ThinkingMoonshot AI94.7%
- GPT-5OpenAI94.3%
- GPT-5.1OpenAI94.0%
- gpt-oss-120bOpenAI93.4%
- GPT-5.1-Codex-MiniOpenAI91.7%
- GPT-5 MiniOpenAI90.7%
- o4 Mini HighOpenAI90.7%
Competition mathematics.
View as table
| Model | Score |
|---|---|
| GPT-5.2 (OpenAI) | 99.0% |
| GPT-5.1-Codex (OpenAI) | 95.7% |
| GLM 4.7 (Z.ai) | 95.0% |
| Kimi K2 Thinking (Moonshot AI) | 94.7% |
| GPT-5 (OpenAI) | 94.3% |
| GPT-5.1 (OpenAI) | 94.0% |
| gpt-oss-120b (OpenAI) | 93.4% |
| GPT-5.1-Codex-Mini (OpenAI) | 91.7% |
| GPT-5 Mini (OpenAI) | 90.7% |
| o4 Mini High (OpenAI) | 90.7% |
MMLU-Pro
Percentage correct · higher is better
- Claude Opus 4.5Anthropic88.9%
- M2.1MiniMax87.5%
- GPT-5.2OpenAI87.4%
- GPT-5OpenAI87.1%
- GPT-5.1OpenAI87.0%
- Gemini 2.5 ProGoogle86.2%
- GPT-5.1-CodexOpenAI86.0%
- Claude Sonnet 4.5Anthropic86.0%
- Claude Opus 4Anthropic86.0%
- GLM 4.7Z.ai85.6%
Broad multi-task knowledge, harder distractors than MMLU.
View as table
| Model | Score |
|---|---|
| Claude Opus 4.5 (Anthropic) | 88.9% |
| M2.1 (MiniMax) | 87.5% |
| GPT-5.2 (OpenAI) | 87.4% |
| GPT-5 (OpenAI) | 87.1% |
| GPT-5.1 (OpenAI) | 87.0% |
| Gemini 2.5 Pro (Google) | 86.2% |
| GPT-5.1-Codex (OpenAI) | 86.0% |
| Claude Sonnet 4.5 (Anthropic) | 86.0% |
| Claude Opus 4 (Anthropic) | 86.0% |
| GLM 4.7 (Z.ai) | 85.6% |
IFBench
Percentage correct · higher is better
- MiniMax M3MiniMax82.9%
- Grok 4.3xAI81.3%
- Grok 4.20xAI81.2%
- Qwen3.7 MaxQwen80.5%
- MiMo-V2.5-ProXiaomi79.9%
- 78.8%
- Qwen3.7 PlusQwen78.0%
- GPT-5.2-CodexOpenAI77.6%
- Gemini 3.1 Flash LiteGoogle77.2%
- 77.2%
Precise instruction following under explicit constraints.
View as table
| Model | Score |
|---|---|
| MiniMax M3 (MiniMax) | 82.9% |
| Grok 4.3 (xAI) | 81.3% |
| Grok 4.20 (xAI) | 81.2% |
| Qwen3.7 Max (Qwen) | 80.5% |
| MiMo-V2.5-Pro (Xiaomi) | 79.9% |
| Qwen3.5 397B A17B (Qwen) | 78.8% |
| Qwen3.7 Plus (Qwen) | 78.0% |
| GPT-5.2-Codex (OpenAI) | 77.6% |
| Gemini 3.1 Flash Lite (Google) | 77.2% |
| Gemini 3.1 Flash Lite Preview (Google) | 77.2% |
Cross-section
The frontier, benchmark by benchmark
The top twelve models on the Intelligence Index, and how each one actually did on the tests behind it.
Frontier models across six evaluations
Percentage correct · darker means stronger
| Model | GPQA Diamond | Humanity's Last Exam | SciCode | τ²-bench | Terminal-Bench Hard | AA-LCR |
|---|---|---|---|---|---|---|
| Claude Opus 5Anthropic | 93% | 55% | 56% | 42% | — | 76% |
| Claude Fable 5Anthropic | 93% | 56% | 60% | 99% | 63% | 77% |
| GPT-5.6 SolOpenAI | 94% | 50% | 56% | 85% | 66% | 78% |
| Kimi K3Moonshot AI | 94% | 47% | 59% | 46% | — | 83% |
| Qwen3.8 MaxQwen | 93% | 43% | 53% | 51% | — | 74% |
| Claude Opus 4.8Anthropic | 92% | 49% | 54% | 94% | 58% | 73% |
| Muse Spark 1.2Meta | 90% | 46% | 56% | 35% | — | 83% |
| GPT-5.6 TerraOpenAI | 93% | 43% | 54% | 86% | 58% | 80% |
| GPT-5.5OpenAI | 94% | 46% | 56% | 94% | 61% | 79% |
| Grok 4.5xAI | 93% | 43% | 54% | 42% | — | 74% |
| Claude Sonnet 5Anthropic | 91% | 41% | 54% | 37% | — | 77% |
| Claude Opus 4.7Anthropic | 91% | 42% | 55% | 89% | 52% | 75% |
0%High
Read across a row to see a model's shape, down a column to see which test separates the field. Blank means not run.
View as table
| Model | GPQA Diamond | Humanity's Last Exam | SciCode | τ²-bench (Telecom) | Terminal-Bench Hard | AA-LCR (long context) |
|---|---|---|---|---|---|---|
| Claude Opus 5 | 93.2% | 54.9% | 55.7% | 42.1% | — | 75.7% |
| Claude Fable 5 | 92.6% | 55.5% | 60.2% | 98.5% | 62.9% | 76.7% |
| GPT-5.6 Sol | 94.1% | 49.5% | 56.1% | 85.1% | 65.9% | 77.7% |
| Kimi K3 | 93.5% | 46.9% | 58.7% | 46.0% | — | 82.7% |
| Qwen3.8 Max | 92.7% | 43.0% | 52.9% | 51.3% | — | 74.3% |
| Claude Opus 4.8 | 92.0% | 48.7% | 53.5% | 94.4% | 58.3% | 73.0% |
| Muse Spark 1.2 | 90.4% | 45.5% | 56.4% | 34.8% | — | 83.3% |
| GPT-5.6 Terra | 92.5% | 42.9% | 53.9% | 86.3% | 57.6% | 79.7% |
| GPT-5.5 | 93.5% | 45.8% | 56.1% | 93.9% | 60.6% | 79.0% |
| Grok 4.5 | 93.1% | 42.7% | 54.1% | 42.1% | — | 74.0% |
| Claude Sonnet 5 | 91.1% | 41.3% | 53.6% | 37.3% | — | 77.0% |
| Claude Opus 4.7 | 91.4% | 42.3% | 54.5% | 88.6% | 51.5% | 75.3% |
Economics
Cost and responsiveness
The two boards where a low number wins.
Cheapest per task
Estimated USD · lower is better
- Ling-3.0-flashInclusionAI$0.006
- V4 Flash 0423DeepSeek$0.02
- V4 Flash 0731DeepSeek$0.02
- Hy3 previewTencent$0.02
- MiMo-V2.5Xiaomi$0.03
- KAT-Coder-Pro V2KwaiPilot$0.04
- Hy3Tencent$0.05
- GPT-5.6 LunaOpenAI$0.05
- Ring-2.6-1TInclusionAI$0.05
- M2.5MiniMax$0.08
Estimated from list pricing: 50K input tokens plus 80K output tokens for reasoning models (25K for non-reasoning). Limited to models scoring 30+ on the Intelligence Index.
View as table
| Model | Cost / task | Intelligence |
|---|---|---|
| Ling-3.0-flash (InclusionAI) | $0.006 | 37.8 |
| V4 Flash 0423 (DeepSeek) | $0.02 | 51.8 |
| V4 Flash 0731 (DeepSeek) | $0.02 | 51.8 |
| Hy3 preview (Tencent) | $0.02 | 34.4 |
| MiMo-V2.5 (Xiaomi) | $0.03 | 38.0 |
| KAT-Coder-Pro V2 (KwaiPilot) | $0.04 | 33.9 |
| Hy3 (Tencent) | $0.05 | 42.2 |
| GPT-5.6 Luna (OpenAI) | $0.05 | 52.3 |
| Ring-2.6-1T (InclusionAI) | $0.05 | 31.1 |
| M2.5 (MiniMax) | $0.08 | 34.5 |
Lowest latency
Seconds to first token · lower is better
- Granite 4.1 8BIBM Granite118ms
- 138ms
- Llama 3 8B LunarisSao10K158ms
- gpt-oss-120bOpenAI205ms
- Qwen3 32BQwen205ms
- Codestral 2508Mistral AI208ms
- 212ms
- MythoMax 13BGryphe222ms
- Hermes 4 70BNous Research229ms
- gpt-oss-safeguard-20bOpenAI237ms
Median time to first token.
View as table
| Model | Latency |
|---|---|
| Granite 4.1 8B (IBM Granite) | 118ms |
| Llama Guard 4 12B (Meta) | 138ms |
| Llama 3 8B Lunaris (Sao10K) | 158ms |
| gpt-oss-120b (OpenAI) | 205ms |
| Qwen3 32B (Qwen) | 205ms |
| Codestral 2508 (Mistral AI) | 208ms |
| Llama 3.2 3B Instruct (Meta) | 212ms |
| MythoMax 13B (Gryphe) | 222ms |
| Hermes 4 70B (Nous Research) | 229ms |
| gpt-oss-safeguard-20b (OpenAI) | 237ms |
323 models considered for every board on this page.