Benchmark Almanac

Terminal-Bench-Science

GPT-6 Astra 68.1% leads since September 2026 Gemini 3.8 Flash 12.4% leads in mini-SWE-agent
GPT-6.2 Astra 77.1% leads since November 2026
In use

The official leaderboard hasn’t scored a model released after Claude Opus 5.5 (September 2026). Labs report their own numbers for 14 newer models, such as GPT-6.2 Astra at 77.1%.

70 scientific research-workflow tasks across five science areas, run in Harbor containers. Hidden tests give the resolution rate.

70 tasks kept from 920 proposals. v0.1. V0.2 in progress. Citation is a Zenodo DOI, no paper.

Released
August 2026
Kind
Research science
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Code → Text, Code, Actions
Interaction
Agentic
Environment
Terminal
Run by
Anyone
Strongest model tested
GPT-6.2 Astra #1 on the capability index
Last in a launch post
November 2026
Last model added
November 2026
Scores refreshed
November 20th, 2026
17
Entries on its own leaderboard, 13 models we track
15
Launch posts list it, from 6 labs
9
System cards report it, from 3 labs

GPT-6 Astra leads with 68.1% among models released by September 2026 GPT-6 Astra leads with 68.1% among models released by September 2026 Gemini 3.8 Flash leads with 12.4% in mini-SWE-agent

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran on its own with several settings, we keep its best. Every run is in the next chart. Agents other companies built around a model get a pale shade of that model's lab color. Gray marks blends and entries not in our model list, such as fine-tunes and older models. None of these count toward leaders. Agents built around a model, such as Codex + DeepSeek V4.1 Flash, are their own entries. We don't count them as the model's score. This view shows only models we track, each at its best. This view keeps only runs in mini-SWE-agent, a harness no model’s maker built. A maker’s own agent, such as Claude Code with a Claude model, is left out.

GPT-6.2 Astra leads with 77.1%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model. Striped bars are claims, and dotted bars are reruns by Epoch AI or another board.

The newest model with a score here came out in September 2026, 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-6 Astra max 68.1%
  2. 2 Claude Opus 5.5 max 63.3%
  3. 3 Claude Fable 5.1 max 40%
  4. 4 Claude Opus 5 max 30%
  5. 5 GPT-5.6 Sol max 22.4%
  6. 6 Claude Fable 5 max 21.4%
  7. 7 Codex + DeepSeek V4.1 Flash max 15.7%
  8. 8 Grok 4.7 xhigh 14.3%
  9. 9 Gemini 3.8 Flash high 12.4%
  10. 10 Claude Opus 4.8 max 10.5%
  11. 11 GPT-5.6 Terra max 8.57%
  12. 12 Claude Code + GLM-5.3 max 8.10%
  13. 13 Claude Code + Kimi K3 max 7.14%
  14. 14 Grok 4.6 xhigh 7.14%
  15. 15 Gemini 3.7 Flash high 5.71%
  16. 16 Codex + DeepSeek-V4-Pro max 3.81%
  17. 17 GPT-5.6 Luna max 3.33%

Source: Official Terminal-Bench-Science leaderboard

Source: Official Terminal-Bench-Science leaderboard

#EntryScoreSettingAgent byDateSource
1 GPT-6 Astra 68.1% max, Codex OpenAI Sep 2026 Official leaderboard
2 Claude Opus 5.5 63.3% max, Claude Code Anthropic Sep 2026 Official leaderboard
3 Claude Fable 5.1 40% max, Claude Code Anthropic Sep 2026 Official leaderboard
4 Claude Opus 5 30% max, Claude Code Anthropic Jul 2026 Official leaderboard
5 GPT-5.6 Sol 22.4% max, Codex OpenAI Jul 2026 Official leaderboard
6 Claude Fable 5 21.4% max, Claude Code Anthropic Jun 2026 Official leaderboard
7 Codex + DeepSeek V4.1 Flash 15.7% max OpenAI Sep 2026 Official leaderboard
8 Grok 4.7 14.3% xhigh, Grok Build xAI Sep 2026 Official leaderboard
9 Gemini 3.8 Flash 12.4% high, mini-SWE-agent SWE-agent Sep 2026 Official leaderboard
10 Claude Opus 4.8 10.5% max, Claude Code Anthropic May 2026 Official leaderboard
11 GPT-5.6 Terra 8.57% max, Codex OpenAI Jul 2026 Official leaderboard
12 Claude Code + GLM-5.3 8.10% max Anthropic Aug 2026 Official leaderboard
13 Claude Code + Kimi K3 7.14% max Anthropic Jul 2026 Official leaderboard
14 Grok 4.6 7.14% xhigh, Grok Build xAI Aug 2026 Official leaderboard
15 Gemini 3.7 Flash 5.71% high, mini-SWE-agent SWE-agent Aug 2026 Official leaderboard
16 Codex + DeepSeek-V4-Pro 3.81% max OpenAI Apr 2026 Official leaderboard
17 GPT-5.6 Luna 3.33% max, Codex OpenAI Jul 2026 Official leaderboard
#ModelScoreSettingAgent byDateSource
1 GPT-6 Astra 68.1% max, Codex OpenAI Sep 2026 Official leaderboard
2 Claude Opus 5.5 63.3% max, Claude Code Anthropic Sep 2026 Official leaderboard
3 Claude Fable 5.1 40% max, Claude Code Anthropic Sep 2026 Official leaderboard
4 Claude Opus 5 30% max, Claude Code Anthropic Jul 2026 Official leaderboard
5 GPT-5.6 Sol 22.4% max, Codex OpenAI Jul 2026 Official leaderboard
6 Claude Fable 5 21.4% max, Claude Code Anthropic Jun 2026 Official leaderboard
7 Grok 4.7 14.3% xhigh, Grok Build xAI Sep 2026 Official leaderboard
8 Gemini 3.8 Flash 12.4% high, mini-SWE-agent SWE-agent Sep 2026 Official leaderboard
9 Claude Opus 4.8 10.5% max, Claude Code Anthropic May 2026 Official leaderboard
10 GPT-5.6 Terra 8.57% max, Codex OpenAI Jul 2026 Official leaderboard
11 Grok 4.6 7.14% xhigh, Grok Build xAI Aug 2026 Official leaderboard
12 Gemini 3.7 Flash 5.71% high, mini-SWE-agent SWE-agent Aug 2026 Official leaderboard
13 GPT-5.6 Luna 3.33% max, Codex OpenAI Jul 2026 Official leaderboard
#ModelScoreSettingAgent byDateSource
1 Gemini 3.8 Flash 12.4% high, mini-SWE-agent SWE-agent Sep 2026 Official leaderboard
2 Gemini 3.7 Flash 5.71% high, mini-SWE-agent SWE-agent Aug 2026 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-6.2 Astra 77.1% Nov 2026 Lab model card
2 Claude Fable 5.2 74.6% max effort Nov 2026 Lab model card
3 Gemini 4 Neon 70.5% Nov 2026 Lab model card
4 GPT-6 Astra 68.1% max, Codex Sep 2026 Official leaderboard
5 GPT-6.2 Sol 68% Nov 2026 Lab model card
6 Gemini 4 Carbon 64.8% Oct 2026 Lab model card
7 Claude Opus 5.5 63.3% max, Claude Code Sep 2026 Official leaderboard
8 Claude Sonnet 5.5 59.9% max effort Sep 2026 Lab model card
9 Gemini 4 Argon 57.6% Oct 2026 Lab model card
10 GPT-6.1 Sol 57% Sep 2026 Lab model card
11 Claude Fable 5.1 52.6% Sep 2026 Lab model card
12 Gemini 4 Hydrogen 51.4% Nov 2026 Lab model card
13 Kimi K3.1 Pro 38.1% Oct 2026 Lab model card
14 Claude Opus 5 30% max, Claude Code Jul 2026 Official leaderboard
15 GPT-5.6 Sol 22.4% max, Codex Jul 2026 Official leaderboard
16 Claude Fable 5 21.4% max, Claude Code Jun 2026 Official leaderboard
17 Claude Haiku 5.5 20.6% max effort Oct 2026 Lab model card
18 Tencent Hy4 19.8% Nov 2026 Lab model card
19 Grok 4.7 14.3% xhigh, Grok Build Sep 2026 Official leaderboard
20 Gemini 3.8 Flash 12.4% high, mini-SWE-agent Sep 2026 Official leaderboard
21 Claude Opus 4.8 10.5% max, Claude Code May 2026 Official leaderboard
22 GPT-5.6 Terra 8.57% max, Codex Jul 2026 Official leaderboard
23 Grok 4.6 7.14% xhigh, Grok Build Aug 2026 Official leaderboard
24 Gemini 3.7 Flash 5.71% high, mini-SWE-agent Aug 2026 Official leaderboard
25 Inkling 2 Small 5.70% Nov 2026 Lab model card
26 GPT-5.6 Luna 3.33% max, Codex Jul 2026 Official leaderboard
27 Inkling 2 Mini 1.40% Nov 2026 Lab model card

Labs report scores for 17 models in their cards. Switch the scores above to rank them together.

The best score, 68.1%, costs $24 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.

OpenAI Anthropic Agents other companies built, in a pale shade of the model’s lab Cost frontier One model at each effort level
0% 20% 40% 60% 80% $1.00 $2.00 $5.00 $10 $20 $50 $100 Cost per task, log scale GPT-6 Astra, max Claude Opus 5.5, max GPT-5.6 Sol, max Codex + DeepSeek V4.1 Flash, max GPT-5.6 Luna, max 0% 20% 40% 60% 80% $1.00 $10 $100 Cost per task, log scale GPT-6 Astra, max Claude Opus 5.5, max

Source: Official Terminal-Bench-Science leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Astra max Codex 68.1% $24
Claude Opus 5.5 max Claude Code 63.3% $23
Claude Fable 5.1 max Claude Code 40% $30
Claude Opus 5 max Claude Code 30% $33
GPT-5.6 Sol max Codex 22.4% $20
Claude Fable 5 max Claude Code 21.4% $68
Codex + DeepSeek V4.1 Flash max Codex 15.7% $1.84
Grok 4.7 xhigh Grok Build 14.3% $15
Gemini 3.8 Flash high mini-SWE-agent 12.4% $5.32
Claude Opus 4.8 max Claude Code 10.5% $28
GPT-5.6 Terra max Codex 8.57% $9.45
Claude Code + GLM-5.3 max Claude Code 8.10% $13
Grok 4.6 xhigh Grok Build 7.14% $16
Claude Code + Kimi K3 max Claude Code 7.14% $7.26
Gemini 3.7 Flash high mini-SWE-agent 5.71% $6.17
Codex + DeepSeek-V4-Pro max Codex 3.81% $5.33
GPT-5.6 Luna max Codex 3.33% $1.82

The best score, 68.1%, costs $24 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.

OpenAI Google DeepMind xAI Anthropic Cost frontier One model at each effort level
0% 20% 40% 60% 80% $1.00 $2.00 $5.00 $10 $20 $50 $100 Cost per task, log scale GPT-6 Astra, max Claude Opus 5.5, max GPT-5.6 Sol, max Grok 4.7, xhigh Gemini 3.8 Flash, high GPT-5.6 Luna, max 0% 20% 40% 60% 80% $1.00 $10 $100 Cost per task, log scale GPT-6 Astra, max Claude Opus 5.5, max

Source: Official Terminal-Bench-Science leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Astra max Codex 68.1% $24
Claude Opus 5.5 max Claude Code 63.3% $23
Claude Fable 5.1 max Claude Code 40% $30
Claude Opus 5 max Claude Code 30% $33
GPT-5.6 Sol max Codex 22.4% $20
Claude Fable 5 max Claude Code 21.4% $68
Grok 4.7 xhigh Grok Build 14.3% $15
Gemini 3.8 Flash high mini-SWE-agent 12.4% $5.32
Claude Opus 4.8 max Claude Code 10.5% $28
GPT-5.6 Terra max Codex 8.57% $9.45
Grok 4.6 xhigh Grok Build 7.14% $16
Gemini 3.7 Flash high mini-SWE-agent 5.71% $6.17
GPT-5.6 Luna max Codex 3.33% $1.82

3 models have held number one on the official leaderboard since its release in August 2026 5 models have held number one with claims included since its release in August 2026

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

Anthropic OpenAI Google DeepMind
Own leaderboard Not tested since Sep 2026
GPT-6 Astra 68.1%
Lab claims 5 leaders

Sources: Official Terminal-Bench-Science leaderboard and labs’ model cards

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 Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GPT-6.2 Astra GPT-6 Astra Claude Fable 5.1 Claude Opus 5 GPT-5.6 Sol Claude Opus 4.8 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GPT-6.2 Astra GPT-6 Astra

Sources: Official Terminal-Bench-Science leaderboard and labs’ model cards

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) GPT-6.2 Astra Gemini 4 Carbon GPT-6 Astra Claude Fable 5.1 Gemini 3.7 Flash 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) GPT-6.2 Astra Gemini 4 Carbon

Sources: Official Terminal-Bench-Science leaderboard and labs’ model cards

6 labs put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Anthropic
OpenAI
Google DeepMind
Moonshot AI
Tencent
Thinking Machines
DateLaunch postScoreAs the post writes it
Nov 18, 2026 Gemini 4 Neon 70.5 Terminal-Bench-Science 0.1
Nov 17, 2026 GPT-6.2 Sol 68 Terminal-Bench-Science 0.1 (GPT-6.2 Sol Max)
Nov 17, 2026 GPT-6.2 Astra 77.1 Terminal-Bench-Science 0.1 (GPT-6.2 Astra Max)
Nov 12, 2026 Gemini 4 Hydrogen 51.4 Terminal-Bench-Science 0.1
Nov 9, 2026 Claude Fable 5.2 74.6 Agentic scientific research Terminal-Bench-Science 0.1, Fable 5.2 with safeguards, max effort (launch table)
Nov 5, 2026 Inkling 2 Small 5.7 Terminal Bench Science v0.1
Nov 4, 2026 Tencent Hy4 19.8 Terminal Bench Science 0.1
Oct 28, 2026 Gemini 4 Carbon 64.8 Terminal-Bench-Science 0.1 (AA)
Oct 26, 2026 GPT-6.1 Terra Terminal-Bench-Science 0.1, max reasoning (chart)
Oct 24, 2026 Kimi K3.1 Pro 38.1 Terminal-Bench-Science 0.1 (Claude Code)
Oct 13, 2026 Gemini 4 Argon 57.6 Terminal-Bench Science 0.1
Sep 29, 2026 GPT-6.1 Sol 57 Terminal-Bench Science 0.1 (GPT-6.1 Sol Max)
Sep 22, 2026 Claude Opus 5.5 58.7 Agentic scientific research Terminal-Bench-Science 0.1
Sep 3, 2026 GPT-6 Astra 64.6 Terminal-Bench Science 0.1
Sep 1, 2026 Claude Fable 5.1 52.6 Agentic scientific research Terminal-Bench-Science 0.1

9 system cards report it, from 3 labs

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Nov 20, 2026 Inkling 2 Mini model card 1.4 Terminal Bench Science v0.1 (mini-swe-agent v2.4.6) Capabilities
Nov 18, 2026 Gemini 4 Neon model card 70.5 Terminal-Bench-Science 0.1 (results table, p.4) Capabilities
Nov 10, 2026 Gemini 4 Carbon model card 64.8 Terminal-Bench-Science 0.1 (AA) (results table) Capabilities
Nov 9, 2026 Claude Fable 5.2 system card 74.6 Terminal-Bench-Science 0.1, safeguards enabled, Claude Code --bare, max effort, no internet access, 10 trials per task. Mythos-only score 75.3% excluded (Section 8.6) Capabilities
Oct 13, 2026 Gemini 4 Argon model card 57.6 Terminal-Bench Science 0.1 (results table, p.5) Capabilities
Oct 7, 2026 Claude Haiku 5.5 system card 20.6 Terminal-Bench-Science 0.1, max effort, safeguards on with no fallback (8.5) Capabilities
Sep 28, 2026 Claude Sonnet 5.5 system card 59.9 Terminal-Bench-Science 0.1, max effort, Claude Code --bare mode (8.6) Capabilities
Sep 22, 2026 Claude Opus 5.5 system card 58.7 Terminal-Bench-Science 0.1, max effort, Claude Code --bare mode, safeguards with fallback, 10 trials per task (Table 8.1.A, 8.6) Capabilities
Sep 1, 2026 Claude Fable 5.1 system card 52.6 Terminal-Bench-Science 0.1, Claude Code --bare mode, max effort, 10 trials per task (Table 8.1.A, 8.7) Capabilities

The lab's number comes out ahead in 1 of 3 pairs that differ

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. Every benchmark.

Model Lab says Board says Gap What differs
Claude Fable 5.1 52.6% Agentic scientific research Terminal-Bench-Science 0.1 40% Leaderboard, max, Claude Code +13 pts Anthropic used Claude Code in bare mode and averaged ten trials per task. The board score averages three trials per task.
Claude Opus 5.5 58.7% Agentic scientific research Terminal-Bench-Science 0.1 63.3% Leaderboard, max, Claude Code −4.6 pts
GPT-6 Astra 64.6% Terminal-Bench Science 0.1 68.1% Leaderboard, max, Codex −3.5 pts

More in Science and reasoning

All 203 in Science and reasoning
  • GPQA Diamond In use 198 graduate-level 'Google-proof' multiple-choice questions in biology, chemistry and physics. Gemini 4 Carbon 97%
  • BioMysteryBench In use Bioinformatics problems on anonymized real datasets that require real analysis with tools. Rubric-graded, with human-solvable and human-difficult splits. Claude Fable 5.2 83.7%
  • FrontierScience In use Models solve physics, chemistry, and biology problems in an Olympiad track graded by short answer and a Research track graded by 10-point rubrics. Intern-S2-397B 87%
  • HiPhO In use Models solve theory problems from 2024 and 2025 physics olympiad exams, some with diagrams. Official marking schemes grade answers and steps, mapped to medal levels. Gemini 3 Pro 78.3%
  • LABBench2 In use 11 biology research sub-tasks (literature, figures, tables, patents, trials and more) with tool access. Macro-averaged accuracy. GPT-5.2 60.9%
  • CritPt v2 In use Models solve unpublished research-level physics challenges written by working physicists, with answers graded automatically. The v2 release revised the public test challenges. GPT-6 Astra 83.1%

Data audit

All 11 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 5 pass
  • Pass Independent runs Scores from the official leaderboard (17 entries), launch posts (14 scores) and system cards (9 scores).
  • Pass Models scored 27 models we track, plus 4 other entries on its leaderboard.
  • Pass Newest score The newest is from November 2026, for Inkling 2 Mini.
  • Pass Frontier models 9 of the top 10 have a score, led by GPT-6.2 Astra (#1).
  • Pass Headroom The best is 77.1%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on terminal-bench-science.ai: 17 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 1 large gap of 3 pairs, each explained in our notes.
  • Pass Known issues No published audit found problems.

How we check