The frontier, remembered
Five years of
machine minds.
Forty consequential releases on one pinned capability scale. Exact DeepSWE scores remain as secondary context for the recent slice.
major releases
2022–2026
Fig. 01
A continuous model history
Scroll the years · 2022 → 2026
2022
-
InstructGPT
Not scored · N/A
Put RLHF-trained instruction following into the default OpenAI API model line.
The exact InstructGPT 175B source row has a blank ECI score.
Pinned Epoch AI snapshot · 27 Aug 2026 -
PaLM
Not scored · N/A
Demonstrated a 540B dense Pathways model with strong few-shot reasoning and code generation.
The exact PaLM 540B source row has a blank ECI score.
Pinned Epoch AI snapshot · 27 Aug 2026 -
OPT-175B
Not scored · N/A
Opened research access to model weights, code, and a training logbook at GPT-3 scale.
The exact OPT-175B source row has a blank ECI score.
Pinned Epoch AI snapshot · 27 Aug 2026 -
BLOOM
Not scored · N/A
Delivered a transparently trained 176B open multilingual model through a global research collaboration.
The exact BLOOM-176B source row has a blank ECI score.
Pinned Epoch AI snapshot · 27 Aug 2026 -
ChatGPT
Not scored · N/A
Turned an RLHF-tuned GPT-3.5 dialogue model into the breakout consumer interface for generative AI.
ChatGPT is a product milestone with no exact scored model row.
Pinned Epoch AI snapshot · 27 Aug 2026
2023
-
LLaMA 65B
ECI 110 · exact 109.71
Showed that smaller, data-efficient foundation models could rival much larger systems.
ECI 110
Exact source value 109.71 · LLaMA-65B
Epoch AI snapshot · 27 Aug 2026 -
Claude
Not scored · N/A
Introduced Anthropic's steerable assistant and API as a durable second frontier-model line.
The snapshot has no exact scored Claude 1 release row.
Pinned Epoch AI snapshot · 27 Aug 2026 -
GPT-4
ECI 126 · exact 126.19
Established a new closed-model frontier with image input and strong professional-benchmark performance.
ECI 126
Exact source value 126.19 · gpt-4-0314
Epoch AI snapshot · 27 Aug 2026 -
Llama 2 70B
ECI 114 · exact 113.57
Moved the Llama family to broadly available weights licensed for research and commercial use.
ECI 114
Exact source value 113.57 · Llama-2-70b-chat-hf
Epoch AI snapshot · 27 Aug 2026 -
Mistral 7B
ECI 112 · exact 111.79
Reset expectations for compact open models with strong results and an Apache 2.0 release.
ECI 112
Exact source value 111.79 · Mistral-7B-v0.1
Epoch AI snapshot · 27 Aug 2026 -
Gemini 1.0 Pro
ECI 117 · exact 116.93
Launched Google's natively multimodal Ultra, Pro, and Nano foundation-model family.
ECI 117
Exact source value 116.93 · gemini-1.0-pro-001
Epoch AI snapshot · 27 Aug 2026
2024
-
Gemini 1.5
Not scored · N/A
Made a one-million-token experimental context window the new long-context benchmark.
The initial February 15 Gemini 1.5 Pro source row has a blank ECI score.
Pinned Epoch AI snapshot · 27 Aug 2026 -
Claude 3 Opus
ECI 127 · exact 126.51
Established the Haiku, Sonnet, and Opus capability-cost tiers still used by Anthropic.
ECI 127
Exact source value 126.51 · claude-3-opus-20240229
Epoch AI snapshot · 27 Aug 2026 -
Llama 3 70B
ECI 123 · exact 122.56
Brought materially stronger 8B and 70B open-weight models to a broad platform ecosystem.
ECI 123
Exact source value 122.56 · Meta-Llama-3-70B
Epoch AI snapshot · 27 Aug 2026 -
GPT-4o
ECI 129 · exact 128.8
Unified text, audio, image, and video interaction in a real-time flagship model.
ECI 129
Exact source value 128.8 · gpt-4o-2024-05-13
Epoch AI snapshot · 27 Aug 2026 -
Claude 3.5 Sonnet
ECI 130 · exact 130.0
Made a mid-tier model the coding and visual-reasoning frontier while introducing Artifacts.
ECI 130
Exact source value 130.0 · claude-3-5-sonnet-20240620
Epoch AI snapshot · 27 Aug 2026 -
OpenAI o1-preview
ECI 136 · exact 135.77
Made test-time reasoning a distinct product and model-scaling dimension.
ECI 136
Exact source value 135.77 · o1-preview-2024-09-12
Epoch AI snapshot · 27 Aug 2026 -
DeepSeek-V3
ECI 133 · exact 133.12
Pushed open mixture-of-experts capability and training efficiency into the frontier conversation.
ECI 133
Exact source value 133.12 · DeepSeek-V3
Epoch AI snapshot · 27 Aug 2026
2025
-
DeepSeek-R1
ECI 140 · exact 139.72
Released a strong reasoning model and distilled variants under an MIT license.
ECI 140
Exact source value 139.72 · DeepSeek-R1
Epoch AI snapshot · 27 Aug 2026 -
Claude 3.7 Sonnet
ECI 142 · exact 141.8
Combined near-instant and extended reasoning in one model and launched alongside Claude Code.
ECI 142
Exact source value 141.8 · claude-3-7-sonnet-20250219
Epoch AI snapshot · 27 Aug 2026 -
Gemini 2.5 Pro
ECI 145 · exact 144.63
Made reasoning native across Google's next Gemini generation with strong coding performance.
ECI 145
Exact source value 144.63 · gemini-2.5-pro-exp-03-25
Epoch AI snapshot · 27 Aug 2026 -
Llama 4 Maverick
ECI 133 · exact 133.03
Moved Meta's open-weight family to native multimodality and mixture-of-experts designs.
ECI 133
Exact source value 133.03 · Llama-4-Maverick-17B-128E-Instruct
Epoch AI snapshot · 27 Aug 2026 -
Qwen3-235B-A22B
ECI 140 · exact 139.64
Brought switchable thinking and non-thinking modes to a broad open-weight model family.
ECI 140
Exact source value 139.64 · qwen3-235b-a22b
Epoch AI snapshot · 27 Aug 2026 -
Claude Opus 4
ECI 143 · exact 143.02
Established Opus 4 and Sonnet 4 as long-running coding and agent-workflow models.
ECI 143
Exact source value 143.02 · claude-opus-4-20250514
Epoch AI snapshot · 27 Aug 2026 -
GPT-5
ECI 150 · exact 150.0
Unified fast responses, deeper reasoning, and routing in OpenAI's next flagship system.
ECI 150
Exact source value 150.0 · gpt-5-2025-08-07_high
Epoch AI snapshot · 27 Aug 2026
2026 · through 26 Aug
-
GPT-5.5
ECI 159 · exact 158.67 · DeepSWE 67%
Advanced agentic coding, computer use, knowledge work, and scientific research in a new frontier release.
ECI 159
Exact source value 158.67 · gpt-5.5_medium
Epoch AI snapshot · 27 Aug 2026 -
DeepSeek V4 Pro
ECI 149 · exact 149.24 · DeepSWE 63%
1.6T open-weight flagship with a one-million-token context window.
ECI 149
Exact source value 149.24 · deepseek-v4-pro_max
Epoch AI snapshot · 27 Aug 2026 -
Claude Opus 4.8
ECI 158 · exact 157.73 · DeepSWE 59%
Reliability-focused Opus upgrade for long-running agent work.
ECI 158
Exact source value 157.73 · claude-opus-4-8_max
Epoch AI snapshot · 27 Aug 2026 -
Claude Fable 5
ECI 162 · exact 162.48 · DeepSWE 70%
New generally available capability frontier with additional safeguards.
ECI 162
Exact source value 162.48 · claude-fable-5
Epoch AI snapshot · 27 Aug 2026 -
Claude Sonnet 5
ECI 156 · exact 155.87 · DeepSWE 54%
Agentic Sonnet release that narrows the gap to the Opus tier.
ECI 156
Exact source value 155.87 · claude-sonnet-5_max
Epoch AI snapshot · 27 Aug 2026 -
GPT-5.6 Luna
ECI 156 · exact 156.21 · DeepSWE 67%
Low-cost GPT-5.6 tier broadens access to strong coding performance.
ECI 156
Exact source value 156.21 · gpt-5.6-luna_max
Epoch AI snapshot · 27 Aug 2026 -
GPT-5.6 Sol
ECI 161 · exact 161.06 · DeepSWE 73%
Flagship GPT-5.6 tier sets the family capability ceiling.
ECI 161
Exact source value 161.06 · gpt-5.6-sol_max
Epoch AI snapshot · 27 Aug 2026 -
Kimi K3
ECI 157 · exact 157.33 · DeepSWE 69%
First open 3T-class multimodal model for long-horizon work.
ECI 157
Exact source value 157.33 · kimi-k3_max
Epoch AI snapshot · 27 Aug 2026 -
Gemini 3.6 Flash
ECI 154 · exact 154.14 · DeepSWE 47%
Efficiency-focused workhorse for production agents.
ECI 154
Exact source value 154.14 · gemini-3.6-flash_high
Epoch AI snapshot · 27 Aug 2026 -
Claude Opus 5
ECI 162 · exact 161.59 · DeepSWE 74%
Generational Opus upgrade for long-running agents.
ECI 162
Exact source value 161.59 · claude-opus-5_max
Epoch AI snapshot · 27 Aug 2026 -
Qwen3.8-Max
Not scored · N/A · DeepSWE 57%
2.4T multimodal flagship for autonomous and long-horizon work.
The official launch is August 3 while Epoch records August 2; the date semantics are unresolved.
Pinned Epoch AI snapshot · 27 Aug 2026 -
Grok 4.6
ECI 156 · exact 156.15 · DeepSWE 67%
Frontier upgrade aimed at long-running and visual agents.
ECI 156
Exact source value 156.15 · grok-4.6_xhigh
Epoch AI snapshot · 27 Aug 2026 -
Gemini 3.7 Flash
ECI 157 · exact 157.12 · DeepSWE 65%
Rapid Flash upgrade focused on coding and agent workflows.
ECI 157
Exact source value 157.12 · gemini-3.7-flash_high
Epoch AI snapshot · 27 Aug 2026 -
GLM-5.3
Not scored · N/A · DeepSWE 69%
Open-weight coding leader improved entirely through post-training.
All exact August 14 GLM-5.3 rows have blank ECI scores.
Pinned Epoch AI snapshot · 27 Aug 2026 -
GLM-5.3-Flash
Not scored · N/A · DeepSWE 63%
Native-multimodal GLM with frontier performance at flash cost.
The release is absent and follows the snapshot's latest scored release date.
Pinned Epoch AI snapshot · 27 Aug 2026
How to read the marks
- 150Filled bubble · rounded ECI point
- Hollow marker · not scored in the pinned ECI snapshot
Circle size is fixed. The ECI axis is visibly truncated to 100–165; DeepSWE appears only as secondary detail where an exact row exists.
How ECI is built · Epoch source snapshotReading note
What the marks mean
Each mark is one release and opens its official evidence. ECI positions the mark; a hollow marker preserves historical context without a fabricated ECI value.
Source: Epoch AI, snapshot retrieved 27 Aug 2026 · CC BY 4.0
ECI combines 50+ benchmarks on an anchored, arbitrary point scale. It is not a percentage or absolute measure of intelligence, and historical values can move when Epoch refits the full dataset.
“Not scored” means no exact scored release-entity row under the pinned snapshot. It is never treated as zero. Exact DeepSWE v1.1 Pass@1 values are retained as secondary context for 15 recent releases, including ECI-N/A late GLM releases.
40 major AI model releases from 2022–2026, through 26 August 2026. Editorial selection, not an exhaustive model catalog.
Read Epoch’s ECI methodology