AIEvals

Models

Inkling

Thinking Machines’ Inkling release, with independently published coding and reasoning results.

Thinking MachinesReleased 15 July 2026Context 256k tokensOutput 256k tokensOpen weights$1.00 in / $4.05 out per Mtok

Family Inkling51 benchmarks with a published resultNo first place on the independent columns among the 46 models this site listsRead 17 October 2026

Where it stands

Rank compares tracked models with a score. Published shows the source’s rank and field size, when available.

BenchmarkScoreRankPublishedLeaderBehind
Vals IndexVals AI28.67%33of 3339of 41Gemini 4 Argon40.23 pts
Artificial Analysis Intelligence IndexArtificial Analysis2541of 42no list readClaude Opus 5.533 pts
Terminal-Bench 4.0Artificial Analysis1.01%39of 40no list readClaude Sonnet 5.562.63 pts
SWE-bench (Verified)Vals AI77.60%21of 2334of 88Claude Opus 519.40 pts
LiveCodeBenchVals AI85.52%14of 2132of 143Claude Fable 5.15.00 pts
GPQA DiamondVals AI87.12%21of 2340of 138Gemini 3.1 Pro8.33 pts
MMLU ProVals AI86.30%20of 2443of 138Claude Fable 5.16.08 pts
MMMU ProVals AI78.50%19of 2051of 93Claude Fable 5.112.14 pts
Vibe Code Bench v1.1Vals AI19.21%37of 3877of 106Claude Sonnet 5.573.18 pts
Terminal-Bench Science 0.1Artificial Analysis0.00%24of 24no list readGPT-6 Astra63.33 pts
ProgramBenchVals AI0.00%19of 2923of 54Claude Opus 5.518.50 pts
τ³-BankingArtificial Analysis29.07%27of 30no list readQwen3.8 Max22.27 pts
AA-AnalystAgentArtificial Analysis23.75%16of 18no list readGemini 3.7 Flash36.25 pts
MLCR-AAArtificial Analysis12.22%34of 37no list readClaude Sonnet 5.562.78 pts
SWE-Bench Pro V2 FullScale AI / SEAL89.88%10of 10no list readClaude Opus 59.52 pts
SWE-Bench Pro V2 HARDScale AI / SEAL56.90%10of 10no list readClaude Opus 541.10 pts
Vibe Code Bench 1–100Vals AI7.33%19of 2019of 20Claude Opus 5.523.03 pts
EnterpriseOps-Gym-AAArtificial Analysis37.96%18of 22no list readClaude Fable 513.16 pts
AutomationBench-AAArtificial Analysis4.98%40of 40no list readGemini 4 Argon72.53 pts
SciCodeArtificial Analysis46.99%39of 40no list readClaude Opus 5.519.91 pts
GDP.pdfArtificial Analysis12.80%34of 40no list readGPT-6.1 Sol18.20 pts
CritPtArtificial Analysis5.43%40of 42no list readGPT-5.6 Sol26.86 pts
AA-OmniscienceArtificial Analysis233of 42no list readClaude Opus 5.544.42
AA-LCR v1.1Artificial Analysis77.33%39of 42no list readKimi K311.34 pts
Terminal-Bench 2.1Vals AI47.57%34of 3563of 75Claude Opus 5.540.07 pts
ProofBench v1.1Vals AI0.00%34of 3445of 46Claude Sonnet 5.5100.00 pts
LegalBenchVals AI82.95%26of 2761of 149Claude Fable 55.61 pts
Harvey's Legal Agent BenchmarkVals AI2.08%32of 3742of 73Muse Spark 1.223.34 pts
SkillsBenchVals AI26.09%20of 2034of 35DeepSeek V4.1 Flash43.71 pts
Vals Multimodal IndexVals AI49.40%16of 1727of 33Claude Fable 524.75 pts
Code MigrationVals AI11.79%36of 3756of 71Claude Sonnet 5.558.04 pts
CyberBench v1.1Vals AI66.79%20of 3126of 43GPT-6 Sol11.19 pts
EMBVals AI38.36%36of 3754of 69Claude Fable 5.138.31 pts
Finance Agent (v2)Vals AI46.60%35of 3747of 73Gemini 4 Argon18.80 pts
IOIVals AI14.94%31of 3136of 38Gemini 4 Argon85.06 pts
Legal Research BenchVals AI28.36%35of 3743of 72Muse Spark 1.3 Max26.93 pts
MedCodeVals AI41.19%31of 3554of 104Claude Opus 522.38 pts
MedScribeVals AI85.41%21of 3626of 106Claude Opus 5.56.02 pts
Public Benefits Bench v1.1Vals AI58.46%26of 2932of 45Claude Opus 518.47 pts
SAGEVals AI36.55%29of 3265of 90Claude Opus 4.719.55 pts
Tax Agent BenchVals AI43.52%35of 3746of 64Claude Fable 5.134.12 pts
GDPval-AA v2.1Artificial Analysis1,063.6740of 42no list readClaude Opus 5.5782.5
AA-Briefcase v1.1Artificial Analysis833.6839of 41no list readClaude Opus 5.5988.17
Humanity's Last ExamArtificial Analysis31.88%41of 42no list readClaude Opus 5.529.47 pts

Nearby on Vals Index

The two models above and below on the Vals Index (Vals AI), on every column where this model has a published number.

ModelVals Index%Artificial Analysis Intelligence IndexindexTerminal-Bench 4.0%SWE-bench%LiveCodeBench%GPQA Diamond%MMLU Pro%MMMU Pro%Vibe Code Bench v1.1%Terminal-Bench Science 0.1%ProgramBench%
MiniMax M3MiniMax36.53%292.02%75.00%82.15%92.68%84.22%81.16%47.57%0.48%–
Gemini 3.1 ProGoogle33.44%304.04%78.80%88.49%95.45%90.99%88.21%32.03%–0.00%
InklingThinking Machines28.67%251.01%77.60%85.52%87.12%86.30%78.50%19.21%0.00%0.00%
Bold marks a figure better than this model’s on the same column, whichever model it belongs to. Ranks cover the models shown from the stated source.

Cost and speed, as published

Additional measurements from Artificial Analysis’ own row. Read 30 September 2026.

As publishedCost per taskOutput speed
Inkling (xhigh)—168.3 tok/s

Results by benchmark

Independent and vendor results retain their own harnesses and effort settings.

BenchmarkIndependent runAs reported by a publisher
τ³-BankingTool use29.07%Artificial Analysis
25.0%tau3 bankingReflection AI · 5 October 2026Rank 5 of 6 in this publisher table
IFBenchInstruction followingno published result
79.8%IFBenchReflection AI · 5 October 2026Rank 3 of 5 in this publisher table
AA-AnalystAgentData analysis23.75%Artificial Analysisnot published
MLCR-AAMedical reasoning12.22%Artificial Analysisnot published
SWE-Bench Pro V2 FullRepository coding89.88%Scale AI / SEALnot published
SWE-Bench Pro V2 HARDRepository coding56.90%Scale AI / SEAL
56.9%SWE Bench Pro v2-HardReflection AI · 5 October 2026Rank 4 of 4 in this publisher table
Vibe Code Bench 1–100Application development7.33%Vals AInot published
EnterpriseOps-Gym-AAEnterprise workflows37.96%Artificial Analysisnot published
SWE-Bench Pro v1Repository codingno published result
54.3%SWE Bench Pro v1Reflection AI · 5 October 2026Rank 4 of 5 in this publisher table
AIME 2026Mathematicsno published result
97.1%AIME 2026Reflection AI · 5 October 2026Rank 3 of 3 in this publisher table
MCP AtlasAgentic toolingno published result
76.0%MCP AtlasReflection AI · 5 October 2026Rank 6 of 7 in this publisher table
BrowseComp (context management)Web researchno published result
77.1%BrowseComp w/ context managementReflection AI · 5 October 2026Rank 3 of 4 in this publisher table
AA-LCR (version not specified)Long-context reasoningno published result
77.3%AA-LCRReflection AI · 5 October 2026Rank 8 of 8 in this publisher table
AA-Omniscience (public split)Factual knowledgeno published result
14.2AA Omniscience (index) - public split only (our results)Reflection AI · 5 October 2026Rank 2 of 5 in this publisher table
AutomationBench-AAWorkflow automation4.98%Artificial Analysisnot published
SciCodeScientific coding46.99%Artificial Analysis
46.1%SciCodeReflection AI · 5 October 2026Rank 6 of 7 in this publisher table
GDP.pdfDocument reasoning12.80%Artificial Analysisnot published
CritPtScientific reasoning5.43%Artificial Analysis
5.4%CriPT AAReflection AI · 5 October 2026Rank 7 of 8 in this publisher table
AA-OmniscienceFactual knowledge2Artificial Analysisnot published
AA-LCR v1.1Long-context reasoning77.33%Artificial Analysisnot published
Terminal-Bench 4.0Agentic coding1.01%Artificial Analysis
0.51%Vals AI
not published
Terminal-Bench 2.1Agentic coding47.57%Vals AI
55.06%Artificial Analysis
63.8%Terminal Bench v2.1Reflection AI · 5 October 2026Rank 7 of 8 in this publisher table
SWE-bench (Verified)Software engineering77.60%Vals AI
77.6%SWEBench VerifiedReflection AI · 5 October 2026Rank 2 of 3 in this publisher table
LiveCodeBenchCode generation85.52%Vals AInot published
GPQA DiamondScience reasoning87.12%Vals AI
87.17%Artificial Analysis
87.2%GPQA DiamondReflection AI · 5 October 2026Rank 7 of 8 in this publisher table
MMLU ProKnowledge86.30%Vals AInot published
MMMU ProMultimodal reasoning78.50%Vals AI
73.47%Artificial Analysis
not published
Vibe Code Bench v1.1Product coding19.21%Vals AInot published
Terminal-Bench Science 0.1Agentic science0.00%Artificial Analysisnot published
ProgramBenchAgentic coding0.00%Vals AInot published
ProofBench v1.1Mathematics0.00%Vals AInot published
LegalBenchProfessional knowledge82.95%Vals AInot published
Harvey's Legal Agent BenchmarkProfessional knowledge2.08%Vals AInot published
SkillsBenchAgentic tooling26.09%Vals AInot published
Vals Multimodal IndexComposite49.40%Vals AInot published
Vals IndexComposite28.67%Vals AInot published
Code MigrationAgentic coding11.79%Vals AInot published
CyberBench v1.1Security66.79%Vals AInot published
EMBFinancial modelling38.36%Vals AInot published
Finance Agent (v2)Financial analysis46.60%Vals AInot published
IOICompetitive programming14.94%Vals AInot published
Legal Research BenchProfessional knowledge28.36%Vals AInot published
MedCodeHealthcare administration41.19%Vals AInot published
MedScribeHealthcare administration85.41%Vals AInot published
Public Benefits Bench v1.1Public services58.46%Vals AInot published
SAGEEducation36.55%Vals AInot published
Tax Agent BenchProfessional knowledge43.52%Vals AInot published
GDPval-AA v2.1Knowledge work1,063.67Artificial Analysisnot published
AA-Briefcase v1.1Knowledge work833.68Artificial Analysisnot published
Humanity's Last ExamMultidisciplinary reasoning31.88%Artificial Analysis
29.7%HLE no toolsReflection AI · 5 October 2026Rank 7 of 8 in this publisher table
Artificial Analysis Intelligence IndexComposite25Artificial Analysisnot published
Independent column: the suite named under each figure, one harness per suite. Publisher column: each lab’s own table, reproduced on the benchmark’s page. Publisher ranks compare reported scores in the same table row only; ties share a rank and unreported cells are excluded.

Source notes

From the linked sources. Read 30 September 2026.

Context sources. Vals AI’s model profile lists a 256k context window. Artificial Analysis records a 1M context window for its measured configuration.

Sources

Read on 17 October 2026.