Kimi K3: Benchmarks, Pricing, and Open-Weights Status
Kimi K3 is Moonshot's 2.8T flagship with 1M context, source-labeled benchmark evidence, $3/$15 API pricing, and open weights still pending until July 27, 2026.
Models
LLM Models
Current AI model guides organized by active, preview, restricted, and historical status, with source-labeled pricing and benchmark evidence.
Use this section to answer two separate questions: what is newest, and what can be used normally today. An announced model can be newer without being generally available.
Active Models
These are the current generally usable lanes in this comparison set.
Claude Sonnet 5
The first Claude cost/performance test before Opus API pricing: 1M context, 128K maximum output, $2/$10 introductory pricing through August 31, then $3/$15. Launch comparisons are vendor evidence; retain Opus for highest-accuracy arbitration.
Claude Opus 4.8
The practical premium Claude baseline for architecture, hard debugging, code review, and final arbitration. Official Claude docs list a 1M context window and $5/$25 input/output pricing; Artificial Analysis reports an Intelligence Index score of 61 at max effort.
GLM-5.2
Z.AI’s current flagship coding model: 1M context, 128K maximum output, $1.40/$4.40 input/output pricing, and Artificial Analysis Index 51. Treat it as a value lane to test, not a proven Opus replacement.
Kimi K3
Moonshot’s newest Kimi flagship: 2.8T vendor-stated parameters, 1M context, native vision, $0.30/$3/$15 cache-hit/input/output pricing, and Artificial Analysis Index 57. Active for API and Kimi product testing, but full weights remain pending until the promised July 27 release.
Kimi K2.7 Code
Moonshot’s cheaper routine Kimi coding API lane, with base and HighSpeed model IDs, 256K context, multimodal input, required thinking mode, and $0.95/$4.00 cache-miss input/output pricing.
Active supporting lanes
- GPT-5.5 — active OpenAI baseline while GPT-5.6 remains in preview.
- MiniMax M3 — 1M-context value coding-agent lane to test.
- Xiaomi MiMo — low-cost long-context and open-weight evaluation lane.
- Gemini 3 Flash — Gemini high-context value lane.
- Gemma 4 — local and open model family.
Preview: GPT-5.6
GPT-5.6 Sol, Terra, and Luna
OpenAI has announced three priced tiers, but access is limited to selected API organizations and Codex workspaces. ChatGPT, public enrollment, stable public limits, and general availability are not part of the June 28 preview.
- Sol: flagship, $5/$30.
- Terra: balanced, $2.50/$15.
- Luna: fastest and lowest-cost, $1/$6.
Use the model guide for selection and pricing, the Sol investigation for agent-safety findings, and the access-gates analysis for policy context.
Guarded and restricted: Fable and Mythos
Claude Fable 5 and Mythos 5
Keep these models out of default routing for different reasons. Anthropic restored Fable globally on native Claude surfaces July 1, but included usage is plan-specific, cloud rollout is incomplete, and safeguards and retention still matter. Mythos remains approved-organization access.
- Fable 5: restored globally on native Claude surfaces; $10/$50 guarded escalation, not a default lane.
- Mythos 5: restored to approved US organizations; not a self-serve production option.
- Default: use Opus 4.8 as the practical premium Claude baseline.
Current Evidence Table
benchmark artifact
Current Model Status, Cost, and Evidence
| Model | Provider | Status | Context | Input price | Output price | Coding signal | Tool-use signal | Benchmark evidence | Speed | Verdict | Sources | Checked |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GPT-5.6 Sol | OpenAI | preview Selected API organizations and Codex workspaces only; no public enrollment and no ChatGPT access during preview. | not published | $5.00 / 1M | $30.00 / 1M | OpenAI reports a new state of the art on Terminal-Bench 2.1; expanded and independent results are pending. | API and Codex preview; max reasoning and ultra multi-agent modes are vendor-documented. |
| OpenAI announced a selected-customer Cerebras preview for July; production latency is not verified. | Restricted-preview evaluation only; keep GPT-5.5 as the active OpenAI baseline. | OpenAI GPT-5.6 launch [archive], OpenAI GPT-5.6 preview access [archive], OpenAI GPT-5.6 Preview system card [archive] | 2026-06-28 |
| GPT-5.6 Terra | OpenAI | preview Same selected-organization API/Codex preview gate as Sol; not in ChatGPT. | not published | $2.50 / 1M | $15.00 / 1M | OpenAI positions Terra as competitive with GPT-5.5; no independent normalized result is verified. | API and Codex preview only. |
| not verified | Restricted-preview balanced lane; listed pricing does not make it a generally usable value route. | OpenAI GPT-5.6 launch [archive], OpenAI GPT-5.6 preview access [archive], OpenAI GPT-5.6 Preview system card [archive] | 2026-06-28 |
| GPT-5.6 Luna | OpenAI | preview Same selected-organization API/Codex preview gate as Sol; not in ChatGPT. | not published | $1.00 / 1M | $6.00 / 1M | OpenAI positions Luna as the fastest and lowest-cost GPT-5.6 tier; independent coding quality is not verified. | API and Codex preview only. |
| Vendor-positioned as fastest; measured production latency is not verified. | Restricted-preview low-cost tier; do not present it as a public budget default. | OpenAI GPT-5.6 launch [archive], OpenAI GPT-5.6 preview access [archive], OpenAI GPT-5.6 Preview system card [archive] | 2026-06-28 |
| GPT-5.5 | OpenAI | active Generally available OpenAI baseline while GPT-5.6 remains in limited preview. | 1.05M API; 400K Codex | $5.00 / 1M | $30.00 / 1M | not verified | not verified | not verified | not verified | Primary coding seat while ChatGPT/Codex limits fit the workload. | OpenAI GPT-5.5 API model page, OpenAI GPT-5.5 ChatGPT limits, Artificial Analysis: GPT-5.5, LMArena leaderboard dataset | 2026-06-28 |
| Claude Sonnet 5 | Anthropic | active Generally available across Claude plans, Claude Code, the Claude API, GitHub Copilot, and supported AWS paths. | 1M | $2.00 / 1M | $10.00 / 1M | Anthropic reports substantial coding and agentic gains over Sonnet 4.6; independent normalized results are pending. | Available in Claude Code and the Claude API; adaptive thinking is on by default. |
| No site-owned normalized latency result is verified. | First Claude cost/performance test before Opus API pricing; retain Opus for highest-accuracy arbitration. | Anthropic Claude Sonnet 5 launch [archive], Claude Sonnet 5 migration guide [archive], GitHub Copilot Claude Sonnet 5 launch [archive], Claude Sonnet 5 on AWS [archive] | 2026-07-01 |
| Claude Opus 4.8 | Anthropic | active Current generally available Opus-tier premium baseline. | 1M | $5.00 / 1M | $25.00 / 1M | Practical premium Claude baseline; AIHackers recommends task-level comparison rather than blanket default routing. | Claude API and Claude-native workflow baseline; third-party routing must follow Anthropic terms. |
| Artificial Analysis measured 57.3 output tokens/s; provider and workload latency vary. | Premium review, architecture, hard-debugging, and final-arbitration lane. | Claude models overview [archive], Claude API pricing [archive], Artificial Analysis: Claude Opus 4.8 [archive], Artificial Analysis Intelligence Index v4.1, LMArena leaderboard dataset, Berkeley Function Calling Leaderboard | 2026-07-18 |
| Claude Fable 5 | Anthropic | active Restored globally on native Claude surfaces July 1; included usage is plan-specific and cloud re-enablement is rolling out. | 1M | $10.00 / 1M | $50.00 / 1M | Anthropic reports frontier launch results; independent reproducible ranking is pending. | Guarded-domain requests can refuse or fall back; verify account behavior before routing. |
| not verified | High-cost guarded escalation only; use Opus 4.8 as the practical Claude premium baseline. | Claude models overview [archive], Claude API pricing [archive], Anthropic Fable 5 and Mythos 5 [archive], Anthropic Fable/Mythos access statement [archive], Anthropic Fable 5 redeployment [archive] | 2026-07-01 |
| Claude Mythos 5 | Anthropic | restricted Restored to a set of approved US organizations; broader Glasswing access remains restricted. | 1M | $10.00 / 1M | $50.00 / 1M | General coding quality is not independently verified for an accessible production route. | Invitation-only research access; account and compliance approval required. |
| not verified | Restricted research context, not a normal production or buying recommendation. | Claude models overview [archive], Claude API pricing [archive], Anthropic Fable 5 and Mythos 5 [archive], Anthropic Fable/Mythos access statement [archive], Anthropic Fable 5 redeployment [archive] | 2026-07-01 |
| Kimi K3 | Moonshot AI | active Active Kimi API/product flagship; full weights promised by July 27, 2026 and not inspectable at this review. | 1M | $3.00 / 1M | $15.00 / 1M | Moonshot reports strong max-reasoning launch-suite coding and agent results; AIHackers repo eval is not verified. | Kimi API and Kimi Code support K3; Kimi docs warn to preserve full assistant history and avoid mid-session model switching. |
| Artificial Analysis reports 62 output tokens/s and flags high verbosity; measure total output cost per accepted task. | Test as Kimi's newest 1M-context frontier-adjacent lane; keep K2.7 Code for cheaper routine Kimi coding until K3 passes local CAR tests. | Kimi K3 launch blog [archive], Kimi K3 quickstart [archive], Kimi K3 API pricing [archive], Kimi current model list [archive], Kimi Code model configuration, Artificial Analysis: Kimi K3 [archive] | 2026-07-18 |
| GLM-5.2 | Z.AI | active Current Z.AI flagship coding model and supported-tool value lane. | 1M | $1.40 / 1M | $4.40 / 1M | Z.AI reports 62.1 on SWE-Bench Pro and 81.0 on Terminal-Bench 2.1. | Supported-tool coding lane; BFCL score not imported. |
| Artificial Analysis flags higher output-token use; measure total cost per successful task. | July value pick to test for supported coding-tool workflows; keep Opus/GPT for final arbitration until local evals pass. | Z.AI GLM-5.2 overview [archive], Z.AI pricing [archive], Artificial Analysis: GLM-5.2 article [archive], Artificial Analysis Intelligence Index v4.1, SWE-bench, Berkeley Function Calling Leaderboard | 2026-06-28 |
| Kimi K2.7 Code | Moonshot AI | active Cheaper routine Kimi coding API lane; HighSpeed is the same model at higher token prices. | 256K | $0.95 / 1M | $4.00 / 1M | Kimi K2.7 Code remains the lower-cost 256K coding lane after K3; independent normalized benchmarks are not imported. | OpenAI-compatible API; thinking mode required in the documented K2.7 Code quickstart. |
| HighSpeed model ID exists at a higher token price; latency not independently measured here. | Cheaper routine Kimi coding API lane when Kimi routing fits and 256K context is enough. | Kimi K2.7 Code quickstart [archive], Kimi K2.7 Code pricing [archive], Kimi Code K2.7 release notes [archive], SWE-bench, Berkeley Function Calling Leaderboard | 2026-06-28 |
Independent, vendor, and site-owned evidence are labeled separately. A preview price, pending weight release, or vendor benchmark does not establish production availability or cost per successful task.
Do not compare scores without matching the benchmark variant and harness. SWE-bench Verified, SWE-Bench Pro, Terminal-Bench, Artificial Analysis, vendor preference tests, and site-owned repository tests answer different questions. Use How to Read AI Benchmarks to turn those results into a shortlist before running your own tasks.
Historical Guides
These pages remain indexable for older integrations, pricing history, and search intent. They are not current recommendations.
- Claude Opus 4.5 — historical Opus reference; use Opus 4.8.
- Claude Sonnet 4.6 — prior daily-production baseline; use Sonnet 5 for current routing.
- Kimi K2.5 — historical and older free-access context; use K3 for latest-Kimi searches and K2.7 Code for cheaper routine coding.
- GLM-5.1 — prior GLM release; use GLM-5.2.
- GLM 4.7 — legacy GLM guide.
Older model names can still be correct when a tool genuinely exposes that version or a dated article describes the period. They should not be relabeled as current leaders.
Selection Shortcuts
| Need | Start with | Escalate or compare with |
|---|---|---|
| Daily Claude production | Sonnet 5 | Opus 4.8 |
| Premium Claude review | Opus 4.8 | Fable only after access and compliance checks |
| Low-cost long-context coding | GLM-5.2 | Opus 4.8 or GPT-5.5 for arbitration |
| Latest Kimi frontier test | Kimi K3 | Opus 4.8 or GPT-5.5 for arbitration |
| Cheaper Kimi coding API | Kimi K2.7 Code | GLM-5.2 when 1M context matters |
| OpenAI production | GPT-5.5 | GPT-5.6 only with confirmed preview access |
| Local or private inference | Gemma 4 | Compare hardware and quantization fit |
See Budget Tier, Mid-Range, Premium, and Smart Spend for workload-specific decisions.
Related links
Last verified: July 18, 2026. Model access, prices, limits, safeguards, open-weight status, and benchmark positions can change independently.
Claude Sonnet 5 Guide
Claude Sonnet 5 pricing, 1M context, API migration changes, tokenizer cost implications, availability, and source-labeled launch evidence.
GPT-5.6 Sol, Terra, and Luna Guide
GPT-5.6 Sol, Terra, and Luna pricing, preview access, benchmark evidence, caching, and deployment caveats as of June 28, 2026.
Claude Opus 4.8 Baseline Guide
Claude Opus 4.8 is the active premium Claude baseline for hard coding review, architecture, and arbitration, with current independent evidence.
GLM-5.2: July Value Coding Pick
GLM-5.2 is the July best value coding model to test: 1M context, AA Index 51, $1.40/$4.40 API pricing, and Opus 4.8 comparison math.
Kimi K2.7 Code: Coding Model Guide
Kimi K2.7 Code is Moonshot's cheaper routine coding lane after K3: 256K context, base/HighSpeed pricing, source-labeled benchmark deltas, and eval caveats.
Xiaomi MiMo Guide: V2.5 API, Pricing & Open Weights
Source-backed guide to Xiaomi MiMo V2.5 API, Token Plan limits, open weights, pricing, hardware reality, privacy caveats, and where it fits.
MiniMax M3: Value Coding Model Guide
MiniMax M3 is a value coding model candidate with 1M context, multimodal input, Token Plan economics, and vendor-reported benchmark strength. Test it before replacing premium lanes.
GLM-5.1: Prior Low-Cost Coding Model Guide
GLM-5.1 is Z.AI's prior low-cost coding model context. Use GLM-5.2 for the current 1M-context GLM coding-model guide.
Gemma 4: Private Local AI From Phone to PC
Gemma 4 explained: official release facts, offline iPhone support for the edge models, and which version to start with on standard local hardware.
Qwen 3.6 Plus Preview: Free 1M-Context Eval on OpenRouter
Qwen 3.6 Plus Preview is currently a free OpenRouter-hosted eval path with a 1M token context window. Here is what is verified, what is still missing, and when value hunters should use it.
Gemini 3 Flash: 78% SWE-bench, $0.50/1M Input, 8x Cheaper Than Claude Opus
Google's Gemini 3 Flash delivers frontier coding performance with $0.50/1M input tokens, 65K output limit, and 1M context. Full benchmarks, pricing breakdown, and when to choose it over Claude Opus 4.5 or Kimi k2.5.
Kimi K2.5 Historical Guide
Historical Kimi K2.5 capabilities, benchmarks, and older free-access routes. Use Kimi K3 for newest-Kimi decisions and K2.7 Code for cheaper coding decisions.
GLM 4.7: Legacy Z.AI Model Guide
Legacy guide for Z.AI GLM 4.7. GLM-5.2 is now the current Z.AI coding-plan model; use this page for historical context and GLM-4.7 fallback routing.
Claude Opus 4.5 Historical Guide
Historical Claude Opus 4.5 guide with current routing notes. Compare Sonnet 5, Opus 4.8, restored Fable 5, and GPT-5.5 before buying premium capacity.