Benchmark Almanac

RSI-Inference Bench

Claude Sonnet 5.5 2,044 leads since October 2026
Not in model cards

Coding agents add model support to inference engines, improve serving speed and fix performance regressions. The harness builds and runs each submission.

Apple reports pass rates for model support and speedups for optimization and repairs. It combines the tracks in an Elo rating.

Made by
Apple
Released
October 2026
Kind
Machine learning engineering
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Code → Code, Actions
Interaction
Agentic
Environment
Terminal
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
In views
Agents
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
October 2026
Scores refreshed
November 20th, 2026
15
Models scored on its own leaderboard

Claude Sonnet 5.5 leads with 2,044

Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

Bars start at 1,000.

  1. 1 Claude Sonnet 5.5 2,044
  2. 2 Claude Opus 5.5 1,903
  3. 3 Claude Fable 5.1 1,716
  4. 4 GPT-6.1 Sol 1,685
  5. 5 GPT-6 Astra 1,676
  6. 6 Gemini 4 Argon 1,676
  7. 7 Claude Opus 5 1,559
  8. 8 GPT-6 Sol 1,448
  9. 9 DeepSeek V4.1 Flash 1017 1,405
  10. 10 MiniMax-M3.1-Flash 1,383
  11. 11 GLM-5.3 1,381
  12. 12 DeepSeek V4.1 Flash 1,295
  13. 13 Claude Opus 4.8 1,259
  14. 14 Gemini 3.8 Flash 1,047
  15. 15 Gemini 3.7 Flash 1,022

Source: Official RSI-Inference Bench leaderboard

#ModelScoreSettingDateSource
1 Claude Sonnet 5.5 2,044 Sep 2026 Official leaderboard
2 Claude Opus 5.5 1,903 Sep 2026 Official leaderboard
3 Claude Fable 5.1 1,716 Sep 2026 Official leaderboard
4 GPT-6.1 Sol 1,685 Sep 2026 Official leaderboard
5 GPT-6 Astra 1,676 Sep 2026 Official leaderboard
6 Gemini 4 Argon 1,676 Sep 2026 Official leaderboard
7 Claude Opus 5 1,559 Jul 2026 Official leaderboard
8 GPT-6 Sol 1,448 Sep 2026 Official leaderboard
9 DeepSeek V4.1 Flash 1017 1,405 Oct 2026 Official leaderboard
10 MiniMax-M3.1-Flash 1,383 Oct 2026 Official leaderboard
11 GLM-5.3 1,381 Aug 2026 Official leaderboard
12 DeepSeek V4.1 Flash 1,295 Sep 2026 Official leaderboard
13 Claude Opus 4.8 1,259 May 2026 Official leaderboard
14 Gemini 3.8 Flash 1,047 Sep 2026 Official leaderboard
15 Gemini 3.7 Flash 1,022 Aug 2026 Official leaderboard

Rankings by part

We read 39 parts from the benchmark’s own leaderboard in 2 groups: day-0 model support and inference optimization. Each ranks models on one part of the test. Claude Sonnet 5.5 leads 21 of them, and 6 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Day-0 Model Support

15 models we track have a score here.

  1. 1 Claude Sonnet 5.5 26.7%
  2. 2 Claude Opus 5.5 13.3%
  3. 3 Claude Fable 5.1 10%
  4. 4 GPT-6 Astra ↑1 10%
  5. 5 GPT-6.1 Sol ↓1 10%
  6. 6 Gemini 4 Argon 6.67%
  7. 7 Claude Opus 4.8 ↑6 0%
  8. 8 Claude Opus 5 ↓1 0%
  9. 9 Gemini 3.7 Flash ↑6 0%
  10. 10 GLM-5.3 ↑1 0%

The best score, 2,044, costs $76 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Anthropic OpenAI DeepSeek Cost frontier One model at each effort level
0 500 1,000 1,500 2,000 2,500 $1.00 $10 $100 $1,000 Cost per task, log scale Claude Sonnet 5.5 Claude Fable 5.1 GPT-6.1 Sol DeepSeek V4.1 Flash 1017 0 500 1,000 1,500 2,000 2,500 $1.00 $10 $100 $1,000 Cost per task, log scale Claude Sonnet 5.5 Claude Fable 5.1

Source: Official RSI-Inference Bench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Sonnet 5.5 2,044 $76 1,674,492
Claude Opus 5.5 1,903 $83 805,686
Claude Fable 5.1 1,716 $71 518,308
GPT-6.1 Sol 1,685 $12 372,113
GPT-6 Astra 1,676 $81 254,553
Gemini 4 Argon 1,676 $68 709,509
Claude Opus 5 1,559 $114 344,039
GPT-6 Sol 1,448 $41 282,499
DeepSeek V4.1 Flash 1017 1,405 $2.11 468,299
MiniMax-M3.1-Flash 1,383 $4.51 685,140
GLM-5.3 1,381 $32 454,765
DeepSeek V4.1 Flash 1,295 $2.20 850,046
Claude Opus 4.8 1,259 $118 622,818
Gemini 3.8 Flash 1,047 $60 431,229
Gemini 3.7 Flash 1,022 $57 325,436

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic
Own leaderboard 1 leader

Source: Official RSI-Inference Bench leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard
1,000 1,250 1,500 1,750 2,000 2,250 2026 1,000 1,250 1,500 1,750 2,000 2,250 2026

Source: Official RSI-Inference Bench leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in AI research

All 164 in AI research
  • PaperBench In use Agent replicates ICML 2024 papers from scratch. LLM judge grades against author-written hierarchical rubrics. Qwen 3.8 Max 93%
  • MLS-Bench In use Agent must invent one algorithmic change in an ML research scaffold whose gain transfers across settings. Scored relative to human baselines. Claude Fable 5.1 50.3%
  • PostTrainBench 1.2 In use CLI agent gets a base model, one H100 and a time budget to post-train it for a target benchmark. Scored on normalized gain. Claude Fable 5.1 44.6%
  • MLE-bench In use Kaggle-style ML engineering competitions. Agent trains models and is scored against competition leaderboards. GPT-6.2 Astra 96.1%
  • PMPP-Hard In use Agents write GPU kernels for 69 tasks that must be correct, and on 48 of them also run within 1.25 times the reference time. The score is the area under the solved-tasks curve over a 10-hour budget, out of 690 task-hours. Kimi K3 0.71
  • KernelBench In use Replace PyTorch operators with faster custom GPU kernels. Scored on correctness and speedup. MiniMax-M3 28.8%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (15 entries).
  • Pass Models scored 15 models we track.
  • Pass Newest score The newest is from October 2026, for MiniMax-M3.1-Flash.
  • Pass Frontier models 6 of the top 10 have a score, led by GPT-6 Astra (#4).
  • Note Headroom Scored as an Elo rating, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on rsi-inference.com: 15 rows for models we track at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check