Benchmark Almanac

GoodAI LTM Benchmark

GPT-4 turbo (snapshot unspecified) 9.20 tops its leaderboard Claude 3 Opus 8.70 leads the models we track
Not in model cards

Conversational agents retain, update, and use information across conversations with long stretches of intervening text. Task-specific evaluators compare answers with expected responses and assign scores from zero to one.

Some tasks use model grading. Versions change scripts and evaluation settings, so memory spans and benchmark versions must accompany reported scores.

Made by
GoodAI
Released
February 2024
Kind
Reasoning over long documents
Level
Everyday
Scored
Checked automatically
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Text
Interaction
Multi-turn
Run by
Anyone
Strongest model tested
Claude 3 Opus #173 on the capability index
Last model added
April 2024
Scores refreshed
November 19th, 2026
9
Entries in its paper, 4 models we track

GPT-4 turbo (snapshot unspecified) tops the leaderboard with 9.20 Claude 3 Opus leads with 8.70 among models released by April 2024

Best score per entry from the benchmark's paper, read on November 19th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Claude 3 Opus, at 8.70. This view shows only models we track, each at its best.

The newest model with a score here came out in April 2024, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-4 turbo (snapshot unspecified) Isolated memory span; input context 16384 tokens; score out of 11 9.20
  2. 2 Claude 3 Opus Isolated memory span; input context 16384 tokens; score out of 11 8.70
  3. 3 Llama 3-70B Isolated memory span; input context 8000 tokens; score out of 11 8.40
  4. 4 GPT-4o-mini (snapshot unspecified) Isolated memory span; input context 16384 tokens; score out of 11 8.10
  5. 5 Gemini 1.5 Pro (snapshot unspecified) 200k memory span; input context 1000000 tokens; score out of 11 7.70
  6. 6 GPT-4o (snapshot unspecified) Isolated memory span; input context 128000 tokens; score out of 11 7.60
  7. 7 Mixtral 8x7B Isolated memory span; input context 32000 tokens; score out of 11 5.00
  8. 8 Mixtral 8x22B Isolated memory span; input context 65536 tokens; score out of 11 4.90
  9. 9 GPT-3.5 turbo (snapshot unspecified) Isolated memory span; input context 16384 tokens; score out of 11 4.10

Source: The GoodAI LTM Benchmark paper

#EntryScoreSettingDateSource
1 GPT-4 turbo (snapshot unspecified) 9.20 Isolated memory span; input context 16384 tokens; score out of 11, LTM Official leaderboard
2 Claude 3 Opus 8.70 Isolated memory span; input context 16384 tokens; score out of 11, LTM Feb 2024 Official leaderboard
3 Llama 3-70B 8.40 Isolated memory span; input context 8000 tokens; score out of 11, LTM Apr 2024 Official leaderboard
4 GPT-4o-mini (snapshot unspecified) 8.10 Isolated memory span; input context 16384 tokens; score out of 11, LTM Official leaderboard
5 Gemini 1.5 Pro (snapshot unspecified) 7.70 200k memory span; input context 1000000 tokens; score out of 11 Official leaderboard
6 GPT-4o (snapshot unspecified) 7.60 Isolated memory span; input context 128000 tokens; score out of 11 Official leaderboard
7 Mixtral 8x7B 5.00 Isolated memory span; input context 32000 tokens; score out of 11 Dec 2023 Official leaderboard
8 Mixtral 8x22B 4.90 Isolated memory span; input context 65536 tokens; score out of 11 Apr 2024 Official leaderboard
9 GPT-3.5 turbo (snapshot unspecified) 4.10 Isolated memory span; input context 16384 tokens; score out of 11 Official leaderboard
#ModelScoreSettingDateSource
1 Claude 3 Opus 8.70 Isolated memory span; input context 16384 tokens; score out of 11, LTM Feb 2024 Official leaderboard
2 Llama 3-70B 8.40 Isolated memory span; input context 8000 tokens; score out of 11, LTM Apr 2024 Official leaderboard
3 Mixtral 8x7B 5.00 Isolated memory span; input context 32000 tokens; score out of 11 Dec 2023 Official leaderboard
4 Mixtral 8x22B 4.90 Isolated memory span; input context 65536 tokens; score out of 11 Apr 2024 Official leaderboard

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude 3 Opus LTM
Isolated me… 32k memory … 200k memory… 500k memory…
8.70
Llama 3-70B LTM
Isolated me… 32k memory …
8.40
Claude 3 Opus
Isolated me… 32k memory … 200k memory… 500k memory…
8.30
Llama 3-70B
Isolated me… 32k memory …
8.20
Mixtral 8x7B
Isolated me… 32k memory …
5.00
Mixtral 8x22B
Isolated me… 32k memory …
4.90

Source: The GoodAI LTM Benchmark paper

ModelEffortHarnessScoreCost per taskOutput tokens
Claude 3 Opus Isolated memory span; input context 16384 tokens; score out of 11 LTM 8.70
Llama 3-70B Isolated memory span; input context 8000 tokens; score out of 11 LTM 8.40
Claude 3 Opus Isolated memory span; input context 200000 tokens; score out of 11 8.30
Llama 3-70B Isolated memory span; input context 8000 tokens; score out of 11 8.20
Claude 3 Opus 120k memory span; input context 200000 tokens; score out of 11 7.40
Claude 3 Opus 32k memory span; input context 200000 tokens; score out of 11 6.70
Claude 3 Opus 200k memory span; input context 16384 tokens; score out of 11 LTM 6.40
Claude 3 Opus 120k memory span; input context 16384 tokens; score out of 11 LTM 5.70
Llama 3-70B 200k memory span; input context 8000 tokens; score out of 11 LTM 5.60
Claude 3 Opus 200k memory span; input context 200000 tokens; score out of 11 5.10
Mixtral 8x7B Isolated memory span; input context 32000 tokens; score out of 11 5.00
Claude 3 Opus 32k memory span; input context 16384 tokens; score out of 11 LTM 5.00
Llama 3-70B 32k memory span; input context 8000 tokens; score out of 11 LTM 5.00
Claude 3 Opus 500k memory span; input context 16384 tokens; score out of 11 LTM 4.90
Mixtral 8x22B Isolated memory span; input context 65536 tokens; score out of 11 4.90
Llama 3-70B 500k memory span; input context 8000 tokens; score out of 11 LTM 4.80
Llama 3-70B 120k memory span; input context 8000 tokens; score out of 11 LTM 4.70
Claude 3 Opus 500k memory span; input context 200000 tokens; score out of 11 3.40
Llama 3-70B 32k memory span; input context 8000 tokens; score out of 11 0.20
Llama 3-70B 120k memory span; input context 8000 tokens; score out of 11 0.20
Llama 3-70B 200k memory span; input context 8000 tokens; score out of 11 0.20
Llama 3-70B 500k memory span; input context 8000 tokens; score out of 11 0.20
Mixtral 8x7B 32k memory span; input context 32000 tokens; score out of 11 0.10
Mixtral 8x7B 120k memory span; input context 32000 tokens; score out of 11 0.10
Mixtral 8x7B 200k memory span; input context 32000 tokens; score out of 11 0.10
Mixtral 8x7B 500k memory span; input context 32000 tokens; score out of 11 0.10
Mixtral 8x22B 120k memory span; input context 65536 tokens; score out of 11 0.10
Mixtral 8x22B 200k memory span; input context 65536 tokens; score out of 11 0.10
Mixtral 8x22B 500k memory span; input context 65536 tokens; score out of 11 0.10
Mixtral 8x22B 32k memory span; input context 65536 tokens; score out of 11 0.00

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude 3 Opus LTM
Isolated me… 32k memory … 200k memory… 500k memory…
8.70
Llama 3-70B LTM
Isolated me… 32k memory …
8.40
Claude 3 Opus
Isolated me… 32k memory … 200k memory… 500k memory…
8.30
Llama 3-70B
Isolated me… 32k memory …
8.20
Mixtral 8x7B
Isolated me… 32k memory …
5.00
Mixtral 8x22B
Isolated me… 32k memory …
4.90

Source: The GoodAI LTM Benchmark paper

ModelEffortHarnessScoreCost per taskOutput tokens
Claude 3 Opus Isolated memory span; input context 16384 tokens; score out of 11 LTM 8.70
Llama 3-70B Isolated memory span; input context 8000 tokens; score out of 11 LTM 8.40
Claude 3 Opus Isolated memory span; input context 200000 tokens; score out of 11 8.30
Llama 3-70B Isolated memory span; input context 8000 tokens; score out of 11 8.20
Claude 3 Opus 120k memory span; input context 200000 tokens; score out of 11 7.40
Claude 3 Opus 32k memory span; input context 200000 tokens; score out of 11 6.70
Claude 3 Opus 200k memory span; input context 16384 tokens; score out of 11 LTM 6.40
Claude 3 Opus 120k memory span; input context 16384 tokens; score out of 11 LTM 5.70
Llama 3-70B 200k memory span; input context 8000 tokens; score out of 11 LTM 5.60
Claude 3 Opus 200k memory span; input context 200000 tokens; score out of 11 5.10
Mixtral 8x7B Isolated memory span; input context 32000 tokens; score out of 11 5.00
Claude 3 Opus 32k memory span; input context 16384 tokens; score out of 11 LTM 5.00
Llama 3-70B 32k memory span; input context 8000 tokens; score out of 11 LTM 5.00
Claude 3 Opus 500k memory span; input context 16384 tokens; score out of 11 LTM 4.90
Mixtral 8x22B Isolated memory span; input context 65536 tokens; score out of 11 4.90
Llama 3-70B 500k memory span; input context 8000 tokens; score out of 11 LTM 4.80
Llama 3-70B 120k memory span; input context 8000 tokens; score out of 11 LTM 4.70
Claude 3 Opus 500k memory span; input context 200000 tokens; score out of 11 3.40
Llama 3-70B 32k memory span; input context 8000 tokens; score out of 11 0.20
Llama 3-70B 120k memory span; input context 8000 tokens; score out of 11 0.20
Llama 3-70B 200k memory span; input context 8000 tokens; score out of 11 0.20
Llama 3-70B 500k memory span; input context 8000 tokens; score out of 11 0.20
Mixtral 8x7B 32k memory span; input context 32000 tokens; score out of 11 0.10
Mixtral 8x7B 120k memory span; input context 32000 tokens; score out of 11 0.10
Mixtral 8x7B 200k memory span; input context 32000 tokens; score out of 11 0.10
Mixtral 8x7B 500k memory span; input context 32000 tokens; score out of 11 0.10
Mixtral 8x22B 120k memory span; input context 65536 tokens; score out of 11 0.10
Mixtral 8x22B 200k memory span; input context 65536 tokens; score out of 11 0.10
Mixtral 8x22B 500k memory span; input context 65536 tokens; score out of 11 0.10
Mixtral 8x22B 32k memory span; input context 65536 tokens; score out of 11 0.00

2 models have held number one in the benchmark’s paper since its release in February 2024

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

Mistral AI Anthropic
Benchmark paper Not tested since Apr 2024

Source: The GoodAI LTM Benchmark paper

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
In the benchmark’s paper
0.00 2.00 4.00 6.00 8.00 10.0 2024 0.00 2.00 4.00 6.00 8.00 10.0 2024

Source: Official GoodAI LTM Benchmark 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 Long context

All 115 in Long context
  • MRCR v2 In use Multi-round coreference: retrieve the n-th of several similar requests hidden in a long synthetic conversation, up to 1M tokens. Muse Spark 1.3 98.5%
  • LongBench v2 In use 503 multiple-choice questions over 8K-2M-word contexts (QA, dialogue, code repos, structured data). Accuracy. Inkling 2 Small 70.6%
  • GraphWalks In use A graph is given as an edge list in long context. The model answers BFS or parent queries, scored by F1. Claude Fable 5 99.96%
  • OpenAI MRCR v2 In use Models read a long synthetic chat that hides 2, 4 or 8 identical writing requests and must reproduce the response to the i-th one. Scored by mean string match ratio per context bin. Gemini 3.7 Flash 89%
  • CL-bench In use Model must learn new knowledge from a supplied long context (rules, procedures, data) and solve tasks. Strict LLM-verified rubrics. GPT-5.4 27.9%
  • AA-LCR 1.1 In use Models answer 100 questions that need reasoning across sets of real documents of about 100K tokens. An LLM equality checker with a fixed system prompt compares each answer with the key. Step 5 Preview 88.3%

Data audit

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in April 2024, for Llama 3-70B. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the benchmark’s paper (9 entries).
  • Pass Models scored 4 models we track, plus 5 other entries on its leaderboard.
  • Warning Newest score The newest came 2 years ago, in April 2024, for Llama 3-70B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Claude 3 Opus (#173).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 30 rows for models we track and 5 other entries at the last read.
  • Pass Last read November 19th, 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